[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_4483":4484,"_4485":-5,"_9":4486,"_4487":4488},"routes/library.video.$slug",{"_9":10,"_11":12,"_13":-5,"_14":15},"data",{"_16":17},"preview",false,"pageNumber","slug","the-mythical-agent-month","routeData",{"_18":19,"_20":21,"_22":23,"_24":25,"_26":27,"_28":29,"_30":31,"_32":-5,"_33":34,"_35":-5,"_36":37,"_38":39,"_40":41,"_42":43,"_44":-5,"_45":46,"_47":48,"_49":50,"_51":52,"_53":-5,"_54":-5,"_14":55,"_56":-5,"_57":58,"_59":60,"_61":62,"_63":64,"_65":66,"_25":67,"_68":69},"_createdAt","2026-10-23T21:54:05Z","_id","27ad23dc-58b9-4366-a3b9-c73d40aedc73","_rev","ge7qSe37EQ3jcYkO7uARxV","_type","video","_updatedAt","2026-10-26T18:00:54Z","access","Public","afterEvent",true,"agenda","collections",[],"cta","endDate","2026-09-25T02:00:00.000Z","excerpt",[4476],"filmedOn","2026-09-24","library",{"_1969":1970,"_1971":1972},"modules","moreFromLibrary",[1960,1961,1962],"outlines",[1952],"posterFrame",{"_24":1949,"_74":1950},"publishDate","2026-10-26T15:35:00.000Z","skill","slideDeck",{"_24":14,"_1824":15},"speakers","speakersWithOrg",[1846],"startDate","2026-09-24T16:30:00.000Z","tags",[1811,1812,1813,1814],"title","The Mythical Agent-Month","transcript",[150,151,152,153,154,155,156,157,158,159,160,161,162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191,192,193,194,195,196,197,198,199,200,201,202,203,204,205,206,207,208,209,210,211],{"_24":73,"_74":75},"seo",{"_70":71,"_72":50},"OGdescription","Wes McKinney, cofounder of Kenn Software, creator of pandas, and co-creator of Apache Arrow, revisits The Mythical Man-Month for the age of agent factories: why coordination is part of the work, and why judgment and taste matter more as agents take on more of the building.","OGimage","mux.video","asset",{"_18":76,"_20":77,"_22":78,"_24":79,"_26":80,"_81":82,"_9":83,"_84":85,"_86":87,"_88":89,"_90":91},"2026-10-23T22:04:46Z","a8573461-f5fa-4369-972e-05f61f47ca51","mG7vmCjaB1vEw3gqUIfzt5","mux.videoAsset","2026-10-23T22:11:03Z","assetId","zucuwWFK1ms9yaHND9uQkhhgYYMqzZvt",{"_92":93,"_94":95,"_96":97,"_98":99,"_100":82,"_101":102,"_103":104,"_105":106,"_107":108,"_109":110,"_111":104,"_112":113,"_114":77,"_115":116,"_117":118,"_119":106,"_86":87,"_120":121,"_122":89,"_123":124},"playbackId","4jMAtMC7uNlI01kBXJ00i6R00SWSPhae005l","status","ready","uploadId","X6HLdMCaXwd3L01G02jqOc5H02XQiujMXME","url","https://stream.mux.com/4jMAtMC7uNlI01kBXJ00i6R00SWSPhae005l","aspect_ratio","16:9","created_at","1792793190","duration",1056.5995,"encoding_tier","smart","id","ingest_type","on_demand_direct_upload","master_access","none","max_resolution_tier","1080p","max_stored_frame_rate",23.976,"max_stored_resolution","UHD","mp4_support","non_standard_input_reasons",{"_148":149},"passthrough","playback_ids",[145],"progress",{"_117":142,"_143":144},"resolution_tier","tracks",[125,126],"upload_id","video_quality","plus",{"_96":127,"_100":135,"_136":137,"_138":139,"_140":31,"_134":141},{"_96":127,"_100":128,"_129":108,"_130":131,"_132":133,"_134":25},1056.5555,"2ChXm7ujyYFEfozv8dDDF1EcnzDlBIDhstZJNV2nMfg","max_frame_rate","max_height",1152,"max_width",2048,"type","02ECQtQBjyXpbEsnqg8AHdp6VvFxlrt6M","max_channel_layout","stereo","max_channels",2,"primary","audio",100,"state","completed",{"_100":85,"_146":147},"policy","public","video_resolution","3840x2160",{"_212":1804,"_24":214,"_215":1805,"_217":1806,"_219":598},{"_212":1774,"_24":214,"_215":1775,"_217":1776,"_219":220},{"_212":1734,"_24":214,"_215":1735,"_217":1736,"_219":220},{"_212":1714,"_24":214,"_215":1715,"_217":1716,"_219":220},{"_212":1684,"_24":214,"_215":1685,"_217":1686,"_219":220},{"_212":1654,"_24":214,"_215":1655,"_217":1656,"_219":220},{"_212":1614,"_24":214,"_215":1615,"_217":1616,"_219":220},{"_212":1594,"_24":214,"_215":1595,"_217":1596,"_219":220},{"_212":1574,"_24":214,"_215":1575,"_217":1576,"_219":220},{"_212":1554,"_24":214,"_215":1555,"_217":1556,"_219":220},{"_212":1544,"_24":214,"_215":1545,"_217":1546,"_219":377},{"_212":1524,"_24":214,"_215":1525,"_217":1526,"_219":220},{"_212":1517,"_24":214,"_215":1518,"_217":1519,"_219":598},{"_212":1487,"_24":214,"_215":1488,"_217":1489,"_219":220},{"_212":1467,"_24":214,"_215":1468,"_217":1469,"_219":220},{"_212":1447,"_24":214,"_215":1448,"_217":1449,"_219":220},{"_212":1417,"_24":214,"_215":1418,"_217":1419,"_219":220},{"_212":1387,"_24":214,"_215":1388,"_217":1389,"_219":220},{"_212":1357,"_24":214,"_215":1358,"_217":1359,"_219":220},{"_212":1347,"_24":214,"_215":1348,"_217":1349,"_219":220},{"_212":1337,"_24":214,"_215":1338,"_217":1339,"_219":377},{"_212":1317,"_24":214,"_215":1318,"_217":1319,"_219":220},{"_212":1287,"_24":214,"_215":1288,"_217":1289,"_219":220},{"_212":1267,"_24":214,"_215":1268,"_217":1269,"_219":220},{"_212":1237,"_24":214,"_215":1238,"_217":1239,"_219":220},{"_212":1230,"_24":214,"_215":1231,"_217":1232,"_219":598},{"_212":1190,"_24":214,"_215":1191,"_217":1192,"_219":220},{"_212":1170,"_24":214,"_215":1171,"_217":1172,"_219":220},{"_212":1140,"_24":214,"_215":1141,"_217":1142,"_219":220},{"_212":1090,"_24":214,"_215":1091,"_217":1092,"_219":220},{"_212":1070,"_24":214,"_215":1071,"_217":1072,"_219":220},{"_212":1030,"_24":214,"_215":1031,"_217":1032,"_219":220},{"_212":980,"_24":214,"_215":981,"_217":982,"_219":220},{"_212":950,"_24":214,"_215":951,"_217":952,"_219":220},{"_212":943,"_24":214,"_215":944,"_217":945,"_219":598},{"_212":893,"_24":214,"_215":894,"_217":895,"_219":220},{"_212":883,"_24":214,"_215":884,"_217":885,"_219":220},{"_212":873,"_24":214,"_215":874,"_217":875,"_219":377},{"_212":853,"_24":214,"_215":854,"_217":855,"_219":220},{"_212":823,"_24":214,"_215":824,"_217":825,"_219":220},{"_212":793,"_24":214,"_215":794,"_217":795,"_219":220},{"_212":773,"_24":214,"_215":774,"_217":775,"_219":220},{"_212":733,"_24":214,"_215":734,"_217":735,"_219":220},{"_212":713,"_24":214,"_215":714,"_217":715,"_219":220},{"_212":673,"_24":214,"_215":674,"_217":675,"_219":220},{"_212":633,"_24":214,"_215":634,"_217":635,"_219":220},{"_212":623,"_24":214,"_215":624,"_217":625,"_219":377},{"_212":603,"_24":214,"_215":604,"_217":605,"_219":220},{"_212":595,"_24":214,"_215":596,"_217":597,"_219":598},{"_212":575,"_24":214,"_215":576,"_217":577,"_219":220},{"_212":545,"_24":214,"_215":546,"_217":547,"_219":220},{"_212":525,"_24":214,"_215":526,"_217":527,"_219":220},{"_212":515,"_24":214,"_215":516,"_217":517,"_219":220},{"_212":485,"_24":214,"_215":486,"_217":487,"_219":220},{"_212":445,"_24":214,"_215":446,"_217":447,"_219":220},{"_212":395,"_24":214,"_215":396,"_217":397,"_219":220},{"_212":374,"_24":214,"_215":375,"_217":376,"_219":377},{"_212":344,"_24":214,"_215":345,"_217":346,"_219":220},{"_212":324,"_24":214,"_215":325,"_217":326,"_219":220},{"_212":284,"_24":214,"_215":285,"_217":286,"_219":220},{"_212":244,"_24":214,"_215":245,"_217":246,"_219":220},{"_212":213,"_24":214,"_215":216,"_217":218,"_219":220},"_key","33b5e1cd19a8","block","children",[229,230,231],"markDefs",[221,222],"style","normal",{"_212":227,"_24":224,"_225":228},{"_212":223,"_24":224,"_225":226},"83dc3c0","timestamp","time","00:17:30.380","73123d9","00:17:25.500",{"_212":241,"_24":233,"_234":242,"_236":243},{"_212":238,"_24":233,"_234":239,"_236":240},{"_212":232,"_24":233,"_234":235,"_236":237},"01934880f439","span","marks",[223],"text","Appreciate it.","da42babd7fc0",[]," ","2ec107c3c303",[227],"So thank you. And look forward to chatting with you later.","af1c85d030e0",[259,260,261,262,263,264,265],[247,248,249,250],{"_212":257,"_24":224,"_225":258},{"_212":255,"_24":224,"_225":256},{"_212":253,"_24":224,"_225":254},{"_212":251,"_24":224,"_225":252},"080be54","00:17:20.460","4987e92","00:17:09.419","9f91ae6","00:17:05.819","5445c06","00:16:56.400",{"_212":281,"_24":233,"_234":282,"_236":283},{"_212":279,"_24":233,"_234":280,"_236":240},{"_212":276,"_24":233,"_234":277,"_236":278},{"_212":274,"_24":233,"_234":275,"_236":240},{"_212":271,"_24":233,"_234":272,"_236":273},{"_212":269,"_24":233,"_234":270,"_236":240},{"_212":266,"_24":233,"_234":267,"_236":268},"eb2e17bfd81f",[251],"that we can delegate all of this labor and walk away from the machine.","22eab78f51e0",[],"af2b4d2eafd9",[253],"I think there's more opportunity than ever to use these models for leverage for our good ideas and to make them a reality, and to reject the narrative","2dfeac7e5a4a",[],"ca9a1c210a8a",[255],"people with experience and taste and judgment and these skills.","b3bd0431dc5c",[],"6636627e1758",[257],"So we have a lot to do here, but, you know, personally I reject the narrative that our jobs are going to be eliminated, and certainly for","35f3cfbca049",[299,300,301,302,303,304,305],[287,288,289,290],{"_212":297,"_24":224,"_225":298},{"_212":295,"_24":224,"_225":296},{"_212":293,"_24":224,"_225":294},{"_212":291,"_24":224,"_225":292},"48c26a3","00:16:51.840","1e410c7","00:16:45.680","5bfce54","00:16:35.120","46aea33","00:16:24.160",{"_212":321,"_24":233,"_234":322,"_236":323},{"_212":319,"_24":233,"_234":320,"_236":240},{"_212":316,"_24":233,"_234":317,"_236":318},{"_212":314,"_24":233,"_234":315,"_236":240},{"_212":311,"_24":233,"_234":312,"_236":313},{"_212":309,"_24":233,"_234":310,"_236":240},{"_212":306,"_24":233,"_234":307,"_236":308},"473b2c8698ec",[291],"We even have got some projects for personal information management.","67bb6b08a138",[],"1c0464399154",[293],"So that includes agent observability, background code verification, task tracking.","1ba60ecebcd0",[],"c5362469e55e",[295],"might lead to. We've also been building a variety of open source projects which are designed for humans to stay in the loop and to be in charge.","a578dead59cd",[],"3285a3589479",[297],"It's clear it's getting better, but, you know, I think Fable things started to really go off the rails and we're presented with these giant walls of text and it became clear these models, they're for each other and not for me, and I worry about what that","371f704f0afe",[333,334,335],[327,328],{"_212":331,"_24":224,"_225":332},{"_212":329,"_24":224,"_225":330},"700eaf4","00:16:20.040","2cb9470","00:16:08.600",{"_212":341,"_24":233,"_234":342,"_236":343},{"_212":339,"_24":233,"_234":340,"_236":240},{"_212":336,"_24":233,"_234":337,"_236":338},"fb5ec7db6a38",[329],"constitution. Even very basic things like communicating with humans.","5d918e95fc2e",[],"891e4260f6f6",[331],"But to me, this is the reality right now, and these are skills that we can't disengage from. On my team we've even codified some of these default agent behaviors into something we call the clanker","b1f9b90a529e",[356,357,358,359,360],[347,348,349],{"_212":354,"_24":224,"_225":355},{"_212":352,"_24":224,"_225":353},{"_212":350,"_24":224,"_225":351},"d6c812e","00:16:04.759","f626d75","00:16:02.759","2521c18","00:15:52.980",{"_212":371,"_24":233,"_234":372,"_236":373},{"_212":369,"_24":233,"_234":370,"_236":240},{"_212":366,"_24":233,"_234":367,"_236":368},{"_212":364,"_24":233,"_234":365,"_236":240},{"_212":361,"_24":233,"_234":362,"_236":363},"8a129ddecf3a",[350],"Again, maybe in a year, maybe in eighteen months, maybe in three years, things will change.","bc9d95419a1d",[],"b28b9b681ff4",[352],"in twenty twenty-six at least without it.","3f1477a3a954",[],"e0bf032d3c6c",[354],"So kind of wrapping up here, I think these are skills that can be practiced, but they're very important, and I think it's very hard to build things that are good","a6e3efd0c15f",[384,385,386],[378,379],"blockquote",{"_212":382,"_24":224,"_225":383},{"_212":380,"_24":224,"_225":381},"1742e31","00:15:49.060","2499405","00:15:41.540",{"_212":392,"_24":233,"_234":393,"_236":394},{"_212":390,"_24":233,"_234":391,"_236":240},{"_212":387,"_24":233,"_234":388,"_236":389},"c9df5d0b09bf",[380],"So now that everyone can code, it's become clear why many shouldn't.","bee4ee08d1a0",[],"d58689dbd69b",[382],"So Paul Ford from the Ftrain blog described it recently in the New York Times that AI is making it easy to do someone else's job badly.","a87e1d67619e",[413,414,415,416,417,418,419,420,421],[398,399,400,401,402],{"_212":411,"_24":224,"_225":412},{"_212":409,"_24":224,"_225":410},{"_212":407,"_24":224,"_225":408},{"_212":405,"_24":224,"_225":406},{"_212":403,"_24":224,"_225":404},"d54db00","00:15:31.880","0f0ede6","00:15:29.320","04309b7","00:15:28.060","acebcbd","00:15:26.860","8122c22","00:15:21.740",{"_212":442,"_24":233,"_234":443,"_236":444},{"_212":440,"_24":233,"_234":441,"_236":240},{"_212":437,"_24":233,"_234":438,"_236":439},{"_212":435,"_24":233,"_234":436,"_236":240},{"_212":432,"_24":233,"_234":433,"_236":434},{"_212":430,"_24":233,"_234":431,"_236":240},{"_212":427,"_24":233,"_234":428,"_236":429},{"_212":425,"_24":233,"_234":426,"_236":240},{"_212":422,"_24":233,"_234":423,"_236":424},"fc36278e9de3",[403],"It's more of the second, and I think we need to work together to remain in control and to remain centaurs and not reverse centaurs.","c7b7c7bfa8d1",[],"aa3b037cf076",[405],"this deployment, this SSH command.","08785836e347",[],"5b17035301f3",[407],"They need us to run","1d6d7c5da5ed",[],"1f9b72f75df3",[409],"They need us to push that button.","fb266bbd7a71",[],"8c47a32416f2",[411],"And babysitting a fleet of agents who are just giving us instructions about things that they can't do.","deca0caa3a8b",[460,461,462,463,464,465,466],[448,449,450,451],{"_212":458,"_24":224,"_225":459},{"_212":456,"_24":224,"_225":457},{"_212":454,"_24":224,"_225":455},{"_212":452,"_24":224,"_225":453},"04950f3","00:15:18.940","335c0da","00:15:13.920","618bbf9","00:15:09.520","468eec1","00:14:57.459",{"_212":482,"_24":233,"_234":483,"_236":484},{"_212":480,"_24":233,"_234":481,"_236":240},{"_212":477,"_24":233,"_234":478,"_236":479},{"_212":475,"_24":233,"_234":476,"_236":240},{"_212":472,"_24":233,"_234":473,"_236":474},{"_212":470,"_24":233,"_234":471,"_236":240},{"_212":467,"_24":233,"_234":468,"_236":469},"9d8c923a0b7f",[452],"A reverse centaur is a machine helped by a human.","a0b17731d9bf",[],"d7413eae792f",[454],"And the idea is that a centaur is a human helped by a machine.","c7b4debc9dc9",[],"dbcc8037b8ef",[456],"but he wrote a book recently called Reverse Centaur's Guide to Life After AI.","ea02ac785de3",[],"93d27f830bf5",[458],"And you're reduced to a state where basically a factory worker clearing paper jams on the factory floor every morning when you get up to check on your goal sessions. I know he's a little bit of a hater, Cory Doctorow,","92246b04edbc",[497,498,499,500,501],[488,489,490],{"_212":495,"_24":224,"_225":496},{"_212":493,"_24":224,"_225":494},{"_212":491,"_24":224,"_225":492},"182ed59","00:14:51.160","fa73938","00:14:49.160","db8a638","00:14:38.160",{"_212":512,"_24":233,"_234":513,"_236":514},{"_212":510,"_24":233,"_234":511,"_236":240},{"_212":507,"_24":233,"_234":508,"_236":509},{"_212":505,"_24":233,"_234":506,"_236":240},{"_212":502,"_24":233,"_234":503,"_236":504},"70969f0052d6",[491],"And can be, you know, not really understanding the inner workings of how it was designed and why it was designed that way.","3e373b6bdc16",[],"79baa5f2e436",[493],"that you don't understand.","b41b55673224",[],"310e0c11ff91",[495],"And so if you don't, you know, exercise this agency, you know, it snowballs to a point where basically you're babysitting a system","d1dfadb61018",[521],[518],{"_212":519,"_24":224,"_225":520},"cc0df5e","00:14:26.020",{"_212":522,"_24":233,"_234":523,"_236":524},"bdf6c174342e",[519],"Models help me get more work done, certainly, but they aren't supplying my reason for getting up and sitting down at the computer to build something and to work until it is good.","b5708235a98a",[534,535,536],[528,529],{"_212":532,"_24":224,"_225":533},{"_212":530,"_24":224,"_225":531},"09c55b2","00:14:16.680","ada04fa","00:14:06.500",{"_212":542,"_24":233,"_234":543,"_236":544},{"_212":540,"_24":233,"_234":541,"_236":240},{"_212":537,"_24":233,"_234":538,"_236":539},"1d07c6e468f5",[530],"tirelessly to bring it to life, it's something you can practice, it's something you can get better at, but it does take a certain type of personality.","158e9afe72e9",[],"fb4a95519dd4",[532],"And so having the will to make things happen, to see the path of, this is the thing I wanna make, this is the thing I wanna do, and then actually working","296f9959f727",[557,558,559,560,561],[548,549,550],{"_212":555,"_24":224,"_225":556},{"_212":553,"_24":224,"_225":554},{"_212":551,"_24":224,"_225":552},"1d8dcff","00:14:01.780","e65d2c3","00:13:50.920","a6b7d6d","00:13:44.839",{"_212":572,"_24":233,"_234":573,"_236":574},{"_212":570,"_24":233,"_234":571,"_236":240},{"_212":567,"_24":233,"_234":568,"_236":569},{"_212":565,"_24":233,"_234":566,"_236":240},{"_212":562,"_24":233,"_234":563,"_236":564},"278e2833e4b3",[551],"and so that means that we have the ability to change it if we want to.","f68ea161413f",[],"1dc6c9677e85",[553],"thank you for Ghostty, that the people that he respects most are the ones who see the world as mutable and that people are responsible for making the world the way that it is,","ddcf84e44843",[],"e8dd3f982dc2",[555],"But I was writing a version of this talk and I saw a tweet from our friend Mitchell Hashimoto,","4666ccabe47a",[584,585,586],[578,579],{"_212":582,"_24":224,"_225":583},{"_212":580,"_24":224,"_225":581},"613ac82","00:13:40.440","b2f4b31","00:13:34.780",{"_212":592,"_24":233,"_234":593,"_236":594},{"_212":590,"_24":233,"_234":591,"_236":240},{"_212":587,"_24":233,"_234":588,"_236":589},"fd3c159855b2",[580],"We call them clankers. I'll talk a little more about that.","d70e897a77c8",[],"a31aff691b72",[582],"On the agency front, what I like to say is that you are the agent, not them.","73b35b825840",[599],[],"h2",{"_212":600,"_24":233,"_234":601,"_236":602},"c1c03a1ef88e",[],"You Are the Agent","850c9c2d49f1",[612,613,614],[606,607],{"_212":610,"_24":224,"_225":611},{"_212":608,"_24":224,"_225":609},"3b14d72","00:13:24.660","df50abf","00:13:17.860",{"_212":620,"_24":233,"_234":621,"_236":622},{"_212":618,"_24":233,"_234":619,"_236":240},{"_212":615,"_24":233,"_234":616,"_236":617},"33c99b18ee0e",[608],"I think they're great. But if you want to build a really delightful user interface, web application, I'm less convinced.","5344e2973789",[],"133367e17e6d",[610],"So, you know, loops are useful for narrow, verified work, and for small, three-line PRs.","182766fa0568",[629],[626],{"_212":627,"_24":224,"_225":628},"cb88b74","00:13:08.040",{"_212":630,"_24":233,"_234":631,"_236":632},"b77c1de03a6b",[627],"And I think anyone that's trying to convince you that you can do that right now, I think in the most cases is either clueless or selling you something.","cab9bc8bb3ac",[648,649,650,651,652,653,654],[636,637,638,639],{"_212":646,"_24":224,"_225":647},{"_212":644,"_24":224,"_225":645},{"_212":642,"_24":224,"_225":643},{"_212":640,"_24":224,"_225":641},"4709c3a","00:12:59.360","5243457","00:12:48.540","fed05fe","00:12:44.140","d53bbb7","00:12:32.620",{"_212":670,"_24":233,"_234":671,"_236":672},{"_212":668,"_24":233,"_234":669,"_236":240},{"_212":665,"_24":233,"_234":666,"_236":667},{"_212":663,"_24":233,"_234":664,"_236":240},{"_212":660,"_24":233,"_234":661,"_236":662},{"_212":658,"_24":233,"_234":659,"_236":240},{"_212":655,"_24":233,"_234":656,"_236":657},"7bd186384da3",[640],"press okay and walk away and get coffee and come back eight hours later and arrive, be presented with a piece of good quality software at the other end.","a7e6e8bf3d20",[],"3a0065c0766b",[642],"there's a deep yearning, I think, in the software ecosystem right now to be able to create these autonomous software factories where we can speak into our gooseneck microphone and","b192b40154d5",[],"127b8efbad8a",[644],"application? Thank you, Codex. But, you know,","b3a1b5a8ceee",[],"77f8a6a1701b",[646],"But there's always a human in the loop to look at it to say, is this ridiculous security theater, over-engineering, building a fortress of solitude for your to-do list","58560a58f711",[688,689,690,691,692,693,694],[676,677,678,679],{"_212":686,"_24":224,"_225":687},{"_212":684,"_24":224,"_225":685},{"_212":682,"_24":224,"_225":683},{"_212":680,"_24":224,"_225":681},"a81c4a4","00:12:24.279","dd15866","00:12:16.140","b92e5a1","00:12:14.140","aaa9703","00:12:06.779",{"_212":710,"_24":233,"_234":711,"_236":712},{"_212":708,"_24":233,"_234":709,"_236":240},{"_212":705,"_24":233,"_234":706,"_236":707},{"_212":703,"_24":233,"_234":704,"_236":240},{"_212":700,"_24":233,"_234":701,"_236":702},{"_212":698,"_24":233,"_234":699,"_236":240},{"_212":695,"_24":233,"_234":696,"_236":697},"d0fe4bb0c584",[680],"And we use agents to review the code that our other agents generate, and that feedback bubbles up and we feed it back into our agent sessions.","1cd36781a8b9",[],"41964e7cf65b",[682],"I do think that some types of loops are useful, but ones that serve the human in the loop.","6f3d72908236",[],"481022a44b7d",[684],"And I wanted to qualify that.","8baab0609b11",[],"10e2a6ce4190",[686],"And I actually went on the record in July on Twitter to say that I think that loops are BS.","1ffd26cd55fa",[722,723,724],[716,717],{"_212":720,"_24":224,"_225":721},{"_212":718,"_24":224,"_225":719},"f9cbc7f","00:11:57.300","1b41068","00:11:46.800",{"_212":730,"_24":233,"_234":731,"_236":732},{"_212":728,"_24":233,"_234":729,"_236":240},{"_212":725,"_24":233,"_234":726,"_236":727},"2126b2bf2488",[718],"are a series of bad decisions that are incoherent, and ending up in the bad slop part of the timeline where you don't want to be.","f4a6be03e121",[],"0a740fe01fb2",[720],"And that is very dangerous, because, you know, it's very tempting, but again, it's increasing the probability that there","8ad60b6a551e",[748,749,750,751,752,753,754],[736,737,738,739],{"_212":746,"_24":224,"_225":747},{"_212":744,"_24":224,"_225":745},{"_212":742,"_24":224,"_225":743},{"_212":740,"_24":224,"_225":741},"3bbdc69","00:11:44.560","7f80776","00:11:42.560","f5a5ca1","00:11:33.360","2f00df2","00:11:27.840",{"_212":770,"_24":233,"_234":771,"_236":772},{"_212":768,"_24":233,"_234":769,"_236":240},{"_212":765,"_24":233,"_234":766,"_236":767},{"_212":763,"_24":233,"_234":764,"_236":240},{"_212":760,"_24":233,"_234":761,"_236":762},{"_212":758,"_24":233,"_234":759,"_236":240},{"_212":755,"_24":233,"_234":756,"_236":757},"a5f4e743ae99",[740],"A is fine. No, B.","35bdc2e2c6cc",[],"af0cd2354671",[742],"I don't know, it's just A.","52d7f5dae542",[],"35a37aefe079",[744],"in a given day, and it's very tempting in many of those moments when you haven't slept well because you've been up all night vibe coding to say, fine, I'll go with the recommendation.","a7a2828aa1b6",[],"9db359b19226",[746],"Now an agentic engineer is being faced with hundreds of decisions in rapid succession","078a9739b8d5",[782,783,784],[776,777],{"_212":780,"_24":224,"_225":781},{"_212":778,"_24":224,"_225":779},"1952d56","00:11:24.400","d34d465","00:11:17.040",{"_212":790,"_24":233,"_234":791,"_236":792},{"_212":788,"_24":233,"_234":789,"_236":240},{"_212":785,"_24":233,"_234":786,"_236":787},"9bf45da48c44",[778],"They might have a meeting, they might have a stand-up to discuss those decisions.","e117c72610e9",[],"7c015acb040d",[780],"And so even the senior software engineers in the past might have to make five critical code decisions, ten decisions in a given day.","a11a61979d9b",[805,806,807,808,809],[796,797,798],{"_212":803,"_24":224,"_225":804},{"_212":801,"_24":224,"_225":802},{"_212":799,"_24":224,"_225":800},"8bba88c","00:11:11.839","d9790bf","00:11:08.080","b6221c9","00:10:57.840",{"_212":820,"_24":233,"_234":821,"_236":822},{"_212":818,"_24":233,"_234":819,"_236":240},{"_212":815,"_24":233,"_234":816,"_236":817},{"_212":813,"_24":233,"_234":814,"_236":240},{"_212":810,"_24":233,"_234":811,"_236":812},"feee966bf8fd",[799],"but they're also able to finish work much more quickly and present you with the next set of decisions to make.","c5b2dd326ec2",[],"a0bd1243d2a6",[801],"Astra 6, tremendously capable models,","2706d5096059",[],"0b2b87159b61",[803],"And this has gotten even more difficult as time has gone on, and I describe the problem as decision fatigue because, especially now with the latest models like Opus 5.5,","7b773d3a5bc2",[835,836,837,838,839],[826,827,828],{"_212":833,"_24":224,"_225":834},{"_212":831,"_24":224,"_225":832},{"_212":829,"_24":224,"_225":830},"8d70a6d","00:10:47.500","d77bce1","00:10:39.740","112ffe3","00:10:29.000",{"_212":850,"_24":233,"_234":851,"_236":852},{"_212":848,"_24":233,"_234":849,"_236":240},{"_212":845,"_24":233,"_234":846,"_236":847},{"_212":843,"_24":233,"_234":844,"_236":240},{"_212":840,"_24":233,"_234":841,"_236":842},"071aa43b7a69",[829],"And if you delegate too many decisions, you're much more likely to end up with slop, almost guaranteed to end up with slop at the end.","33d4624260ce",[],"04ec4ef6a540",[831],"and no leaf in this infinitely, exponentially large decision tree yields the same product.","13ae4e4afa7d",[],"cdcc85207efb",[833],"And the more of these decisions that you're willing to delegate to agents, the higher the chances that one of those decisions will be bad or suboptimal,","ec7d0b885bd5",[862,863,864],[856,857],{"_212":860,"_24":224,"_225":861},{"_212":858,"_24":224,"_225":859},"0cfbf08","00:10:19.040","cf62ac0","00:10:12.100",{"_212":870,"_24":233,"_234":871,"_236":872},{"_212":868,"_24":233,"_234":869,"_236":240},{"_212":865,"_24":233,"_234":866,"_236":867},"e05914118280",[858],"But if you think about a successful software product, something you charge people money for, there are hundreds, thousands of decisions that are made at any given time.","064d55fad3d3",[],"b6659709b938",[860],"Obviously you can see how this compounds and it snowballs and becomes a liability for the code base.","326c0b4ce5aa",[879],[876],{"_212":877,"_24":224,"_225":878},"8394ec2","00:10:06.560",{"_212":880,"_24":233,"_234":881,"_236":882},"1f6ba6e0d932",[877],"So just because the agent can build it doesn't mean it's right for your product.","8e254e25b134",[889],[886],{"_212":887,"_24":224,"_225":888},"c719284","00:09:57.920",{"_212":890,"_24":233,"_234":891,"_236":892},"700334e3f1ab",[887],"I find that this class of issue occurs acutely in UI because there's a lot of training data and web applications often have sign-out buttons.","edf0c065d79b",[911,912,913,914,915,916,917,918,919],[896,897,898,899,900],{"_212":909,"_24":224,"_225":910},{"_212":907,"_24":224,"_225":908},{"_212":905,"_24":224,"_225":906},{"_212":903,"_24":224,"_225":904},{"_212":901,"_24":224,"_225":902},"ffda8eb","00:09:50.980","0029121","00:09:49.060","1629067","00:09:46.420","840d875","00:09:42.399","f0796d8","00:09:30.580",{"_212":940,"_24":233,"_234":941,"_236":942},{"_212":938,"_24":233,"_234":939,"_236":240},{"_212":935,"_24":233,"_234":936,"_236":937},{"_212":933,"_24":233,"_234":934,"_236":240},{"_212":930,"_24":233,"_234":931,"_236":932},{"_212":928,"_24":233,"_234":929,"_236":240},{"_212":925,"_24":233,"_234":926,"_236":927},{"_212":923,"_24":233,"_234":924,"_236":240},{"_212":920,"_24":233,"_234":921,"_236":922},"c437128d3a1f",[901],"And the agent said, \"Well, I invented a convention that does not exist in your products.\"","4d564d18fc91",[],"539df1283d44",[903],"Life? What is there to sign out of?\"","230125122859",[],"952b5ff7d7f2",[905],"So I said, \"Sign out of what exactly?","349c72c8385e",[],"a94cb647d804",[907],"And it was the type of thing where there was nothing to sign out of.","e6cf1af139d1",[],"c59c674a8f02",[909],"The taste thing is more subtle, I will say. I was working on a product, a web application, and the agent added a sign-out button that I never asked for.","230b843bd818",[946],[],{"_212":947,"_24":233,"_234":948,"_236":949},"93dede62c5ea",[],"Taste and Decision Fatigue","50747ba09944",[962,963,964,965,966],[953,954,955],{"_212":960,"_24":224,"_225":961},{"_212":958,"_24":224,"_225":959},{"_212":956,"_24":224,"_225":957},"48e2162","00:09:19.620","f91cbe4","00:09:13.060","22b4d34","00:09:04.220",{"_212":977,"_24":233,"_234":978,"_236":979},{"_212":975,"_24":233,"_234":976,"_236":240},{"_212":972,"_24":233,"_234":973,"_236":974},{"_212":970,"_24":233,"_234":971,"_236":240},{"_212":967,"_24":233,"_234":968,"_236":969},"9e89bab8181e",[956],"these things can fly by and compound and create, you know, really, really serious problems as the agents find themselves stumbling over the messes that they create inadvertently in the course of fulfilling your wishes.","a46c68dbe6ad",[],"4df055ae6e02",[958],"we're still, if you don't pay attention and really dial in and engage,","65a139da2140",[],"347bc678fdf9",[960],"And, you know, as much as the frontier labs work on addressing this hallucination problem or communication problem, whatever it is,","593dcb54ba9d",[998,999,1000,1001,1002,1003,1004,1005,1006],[983,984,985,986,987],{"_212":996,"_24":224,"_225":997},{"_212":994,"_24":224,"_225":995},{"_212":992,"_24":224,"_225":993},{"_212":990,"_24":224,"_225":991},{"_212":988,"_24":224,"_225":989},"ca405aa","00:08:57.760","a41759f","00:08:54.800","a29eeab","00:08:51.600","75c6a2b","00:08:46.700","c1d2c9e","00:08:43.260",{"_212":1027,"_24":233,"_234":1028,"_236":1029},{"_212":1025,"_24":233,"_234":1026,"_236":240},{"_212":1022,"_24":233,"_234":1023,"_236":1024},{"_212":1020,"_24":233,"_234":1021,"_236":240},{"_212":1017,"_24":233,"_234":1018,"_236":1019},{"_212":1015,"_24":233,"_234":1016,"_236":240},{"_212":1012,"_24":233,"_234":1013,"_236":1014},{"_212":1010,"_24":233,"_234":1011,"_236":240},{"_212":1007,"_24":233,"_234":1008,"_236":1009},"386bb52b09ae",[988],"Exit code 0.\" So a confident summary is not evidence.","a1e71580229e",[],"f4616de03730",[990],"Agent confidently stated, \"Journey 1 completed successfully.","7da11d325162",[],"845a84963384",[992],"\"Exit error 1. Command did not start or failed.\"","53d8388cea00",[],"80b5e4d59349",[994],"One of my colleagues, so on the left we had, you know, a build that failed.","9778b827a021",[],"d00dbcb8536c",[996],"Here's another.","75c996bca5b4",[1045,1046,1047,1048,1049,1050,1051],[1033,1034,1035,1036],{"_212":1043,"_24":224,"_225":1044},{"_212":1041,"_24":224,"_225":1042},{"_212":1039,"_24":224,"_225":1040},{"_212":1037,"_24":224,"_225":1038},"43beede","00:08:37.919","77c0dcb","00:08:28.240","112dbf4","00:08:25.800","efca7ce","00:08:14.520",{"_212":1067,"_24":233,"_234":1068,"_236":1069},{"_212":1065,"_24":233,"_234":1066,"_236":240},{"_212":1062,"_24":233,"_234":1063,"_236":1064},{"_212":1060,"_24":233,"_234":1061,"_236":240},{"_212":1057,"_24":233,"_234":1058,"_236":1059},{"_212":1055,"_24":233,"_234":1056,"_236":240},{"_212":1052,"_24":233,"_234":1053,"_236":1054},"0091c00463a0",[1037],"And if you're not reading the code, you may not even notice that this happened.","1c13695f09ae",[],"6afda15adf8e",[1039],"And so this constraint that was generated by the agent that had no reason to exist in the first place turned into a real constraint that the agents have to contend with.","98d651a25e1f",[],"2c2dfe146b0d",[1041],"to be able to test that database migration.","803aab5850da",[],"228c630820b2",[1043],"The next agent that works on that code base is now responsible for the database migration, the testing, the fixtures, the harnesses that were created, often in a very over-engineered way,","a522952c426e",[1079,1080,1081],[1073,1074],{"_212":1077,"_24":224,"_225":1078},{"_212":1075,"_24":224,"_225":1076},"7d63c16","00:08:12.280","f6ffb9f","00:08:00.560",{"_212":1087,"_24":233,"_234":1088,"_236":1089},{"_212":1085,"_24":233,"_234":1086,"_236":240},{"_212":1082,"_24":233,"_234":1083,"_236":1084},"a9eb902aa113",[1075],"Agents are happy to do that.","624d5a2ca99e",[],"0fe163de9989",[1077],"But you think about this category of problem and how the issue can compound is that if the agent writes a database migration that you don't need, that code that you generate was very cheap to generate.","c231dfae4c29",[1108,1109,1110,1111,1112,1113,1114,1115,1116],[1093,1094,1095,1096,1097],{"_212":1106,"_24":224,"_225":1107},{"_212":1104,"_24":224,"_225":1105},{"_212":1102,"_24":224,"_225":1103},{"_212":1100,"_24":224,"_225":1101},{"_212":1098,"_24":224,"_225":1099},"7ddde62","00:07:58.880","f4d1d28","00:07:55.920","f4f82c7","00:07:52.120","0843d0f","00:07:46.039","b2db956","00:07:37.760",{"_212":1137,"_24":233,"_234":1138,"_236":1139},{"_212":1135,"_24":233,"_234":1136,"_236":240},{"_212":1132,"_24":233,"_234":1133,"_236":1134},{"_212":1130,"_24":233,"_234":1131,"_236":240},{"_212":1127,"_24":233,"_234":1128,"_236":1129},{"_212":1125,"_24":233,"_234":1126,"_236":240},{"_212":1122,"_24":233,"_234":1123,"_236":1124},{"_212":1120,"_24":233,"_234":1121,"_236":240},{"_212":1117,"_24":233,"_234":1118,"_236":1119},"864fc88c110e",[1098],"I don't know.","e02200e9031a",[],"0a0bd397d036",[1100],"Maybe some people are and they live like animals.","095a89fbdf63",[],"af2906bd3580",[1102],"And so my comment was, well, it acts like every commit it's deploying to prod.","b111c9e6ea53",[],"af403ea532a4",[1104],"for those database migrations, without stopping to ask, should we be writing database migrations in the first place?","a3b63a5d180a",[],"ef7b6906baaf",[1106],"And so obviously the agent knows that whenever it changes the schema, it has to write database migrations and it has to write tests","3ecad7e15ad7",[1152,1153,1154,1155,1156],[1143,1144,1145],{"_212":1150,"_24":224,"_225":1151},{"_212":1148,"_24":224,"_225":1149},{"_212":1146,"_24":224,"_225":1147},"31694ae","00:07:28.940","1794ede","00:07:24.699","a612c48","00:07:19.480",{"_212":1167,"_24":233,"_234":1168,"_236":1169},{"_212":1165,"_24":233,"_234":1166,"_236":240},{"_212":1162,"_24":233,"_234":1163,"_236":1164},{"_212":1160,"_24":233,"_234":1161,"_236":240},{"_212":1157,"_24":233,"_234":1158,"_236":1159},"c13ae61b9e78",[1146],"was inserting database migrations for features that were part of a pull request that had never been merged.","dc58bc5c064d",[],"7d148506f4e1",[1148],"And the agent, I think it was Sol, I think it was GPT-5.6 Sol,","2ef5850f6ba5",[],"d11db4051bbb",[1150],"So just a week or two ago I noticed that I was inside of a pull request.","9466b202d9ba",[1179,1180,1181],[1173,1174],{"_212":1177,"_24":224,"_225":1178},{"_212":1175,"_24":224,"_225":1176},"dd158bc","00:07:11.940","490ecb5","00:07:00.580",{"_212":1187,"_24":233,"_234":1188,"_236":1189},{"_212":1185,"_24":233,"_234":1186,"_236":240},{"_212":1182,"_24":233,"_234":1183,"_236":1184},"a605c1dea332",[1175],"showing how this can go off the rails, starting with the judgment piece.","05ed9fd4f40a",[],"c5af2d958c00",[1177],"So rather than kind of explaining these in technical detail, I wanted to give examples, real examples from very recent agent work","e6036681f86e",[1205,1206,1207,1208,1209,1210,1211],[1193,1194,1195,1196],{"_212":1203,"_24":224,"_225":1204},{"_212":1201,"_24":224,"_225":1202},{"_212":1199,"_24":224,"_225":1200},{"_212":1197,"_24":224,"_225":1198},"6835627","00:06:53.139","3dfc176","00:06:47.460","67fd337","00:06:40.420","2c2a536","00:06:28.860",{"_212":1227,"_24":233,"_234":1228,"_236":1229},{"_212":1225,"_24":233,"_234":1226,"_236":240},{"_212":1222,"_24":233,"_234":1223,"_236":1224},{"_212":1220,"_24":233,"_234":1221,"_236":240},{"_212":1217,"_24":233,"_234":1218,"_236":1219},{"_212":1215,"_24":233,"_234":1216,"_236":240},{"_212":1212,"_24":233,"_234":1213,"_236":1214},"5e9baec649dd",[1197],"And then being the person who wills it into existence and makes it happen.","a2b4b636b647",[],"64d9e80cf58e",[1199],"real, which problems are imagined, knowing when you're making something good.","12f4b70b026a",[],"71de528cb6da",[1201],"of judgment, taste, and agency. Certainly deciding what's worth doing, which problems are","669929a3fcc9",[],"ee16889d835d",[1203],"What I concluded back in February, and I believe even more strongly now, is that the thing that the agents are having a hard time supplying for us are these three skills that we can develop and we can practice","f5401f66efcf",[1233],[],{"_212":1234,"_24":233,"_234":1235,"_236":1236},"44656827673c",[],"Judgment, Taste, and Agency","49d0785d923a",[1249,1250,1251,1252,1253],[1240,1241,1242],{"_212":1247,"_24":224,"_225":1248},{"_212":1245,"_24":224,"_225":1246},{"_212":1243,"_24":224,"_225":1244},"8e8b53f","00:06:16.440","dc9d73b","00:06:09.520","978477c","00:06:03.760",{"_212":1264,"_24":233,"_234":1265,"_236":1266},{"_212":1262,"_24":233,"_234":1263,"_236":240},{"_212":1259,"_24":233,"_234":1260,"_236":1261},{"_212":1257,"_24":233,"_234":1258,"_236":240},{"_212":1254,"_24":233,"_234":1255,"_236":1256},"bac46ebc64fd",[1243],"But the thing that you won't necessarily reach is a product that people love, that people want to pay for, that people want to use every day.","46061dcbd77b",[],"17362749541e",[1245],"The harnesses are good at making code that works and that is not, you know, totally ugly.","06e06bb7e3bf",[],"2c79737dc579",[1247],"But insidiously, managing your agent fleet won't necessarily show up as bad code.","b7cd5e0f02e0",[1276,1277,1278],[1270,1271],{"_212":1274,"_24":224,"_225":1275},{"_212":1272,"_24":224,"_225":1273},"6b7a8c2","00:05:57.400","2cff199","00:05:51.560",{"_212":1284,"_24":233,"_234":1285,"_236":1286},{"_212":1282,"_24":233,"_234":1283,"_236":240},{"_212":1279,"_24":233,"_234":1280,"_236":1281},"6900a5789f32",[1272],"So if you're headed towards something that is a disaster, you'll reach the disaster much more quickly.","ee29e9dc608f",[],"a3974f52bb3f",[1274],"I think one founder described it to me that it won't necessarily be better, but just a lot faster.","3b6c145799a4",[1299,1300,1301,1302,1303],[1290,1291,1292],{"_212":1297,"_24":224,"_225":1298},{"_212":1295,"_24":224,"_225":1296},{"_212":1293,"_24":224,"_225":1294},"dd022c8","00:05:46.980","33c2055","00:05:41.940","81a6c5e","00:05:38.420",{"_212":1314,"_24":233,"_234":1315,"_236":1316},{"_212":1312,"_24":233,"_234":1313,"_236":240},{"_212":1309,"_24":233,"_234":1310,"_236":1311},{"_212":1307,"_24":233,"_234":1308,"_236":240},{"_212":1304,"_24":233,"_234":1305,"_236":1306},"3681e1784787",[1293],"And so what's to say that they're going to be able to run an agent factory with 50 agents?","f1e94a448a49",[],"bedf548c1009",[1295],"And so sometimes I'll talk to first-time founders who've never even run a team of five people.","a6f6bdc2cd8d",[],"b3442b63847a",[1297],"And I'm an investor, so I work with a lot of founders.","52997ab3ce89",[1326,1327,1328],[1320,1321],{"_212":1324,"_24":224,"_225":1325},{"_212":1322,"_24":224,"_225":1323},"1ddd31a","00:05:28.500","33de537","00:05:24.180",{"_212":1334,"_24":233,"_234":1335,"_236":1336},{"_212":1332,"_24":233,"_234":1333,"_236":240},{"_212":1329,"_24":233,"_234":1330,"_236":1331},"52626f2fa5d3",[1322],"And so it kind of begs the question, if you have not managed people, what's to say that you're going to be successful at managing fleets of agents to build software?","bb32dc9bd92f",[],"78ba915cf3af",[1324],"And so coordinating agents, it is management.","35de2a9f7d4b",[1343],[1340],{"_212":1341,"_24":224,"_225":1342},"dfee889","00:05:15.980",{"_212":1344,"_24":233,"_234":1345,"_236":1346},"3e4471da86b5",[1341],"But the pushback goes away. So the agents are much more willing to follow a bad idea rather than questioning it.","6b1f70169056",[1353],[1350],{"_212":1351,"_24":224,"_225":1352},"5a23d35","00:05:05.159",{"_212":1354,"_24":233,"_234":1355,"_236":1356},"f3aba8786c00",[1351],"So if you swap agents in and get rid of the humans, you still have the coordination problem of getting different agents working on different parts of the project to march to the beat of the same drum.","57704eb2dfc9",[1369,1370,1371,1372,1373],[1360,1361,1362],{"_212":1367,"_24":224,"_225":1368},{"_212":1365,"_24":224,"_225":1366},{"_212":1363,"_24":224,"_225":1364},"97e3d68","00:04:58.039","62e85c3","00:04:53.160","31efe9f","00:04:43.960",{"_212":1384,"_24":233,"_234":1385,"_236":1386},{"_212":1382,"_24":233,"_234":1383,"_236":240},{"_212":1379,"_24":233,"_234":1380,"_236":1381},{"_212":1377,"_24":233,"_234":1378,"_236":240},{"_212":1374,"_24":233,"_234":1375,"_236":1376},"2931d4ff3827",[1363],"and you hand it to a person and they say, wait, these things don't all add up, this part of the spec isn't coherent with this part of the spec.","ac0d7efd8b92",[],"4209aca90a4f",[1365],"It can be annoying in the moment, but it's tremendously useful if you've written a specification","4d724f87657b",[],"dc21f3cdaad1",[1367],"And one of the very valuable roles that I underestimated with the team is that humans can push back on design that doesn't make sense.","a219ac6dbb5f",[1399,1400,1401,1402,1403],[1390,1391,1392],{"_212":1397,"_24":224,"_225":1398},{"_212":1395,"_24":224,"_225":1396},{"_212":1393,"_24":224,"_225":1394},"d50612d","00:04:36.880","96d43d0","00:04:32.320","099a566","00:04:24.640",{"_212":1414,"_24":233,"_234":1415,"_236":1416},{"_212":1412,"_24":233,"_234":1413,"_236":240},{"_212":1409,"_24":233,"_234":1410,"_236":1411},{"_212":1407,"_24":233,"_234":1408,"_236":240},{"_212":1404,"_24":233,"_234":1405,"_236":1406},"eb75d4aed021",[1393],"So typically you have a software architect and then a team of skilled implementers that support them.","ec41008bdcdd",[],"4591c87388bd",[1395],"And so in The Mythical Man-Month, there's this idea of the surgical team.","be0bc94708a4",[],"8a7d73fae0d0",[1397],"I think what I underweighted back then in February was the team itself.","387fee24e8da",[1429,1430,1431,1432,1433],[1420,1421,1422],{"_212":1427,"_24":224,"_225":1428},{"_212":1425,"_24":224,"_225":1426},{"_212":1423,"_24":224,"_225":1424},"41dcc90","00:04:14.460","338215f","00:04:08.140","f079c43","00:03:57.200",{"_212":1444,"_24":233,"_234":1445,"_236":1446},{"_212":1442,"_24":233,"_234":1443,"_236":240},{"_212":1439,"_24":233,"_234":1440,"_236":1441},{"_212":1437,"_24":233,"_234":1438,"_236":240},{"_212":1434,"_24":233,"_234":1435,"_236":1436},"26be1f45f1e6",[1423],"But there's still a lot of difficult-to-see work involved with taking a working piece of software and turning that into a great software product.","f350a4b7e0ae",[],"52375a0c1222",[1425],"And so I think agents have given us the feeling and the ability of being able to create working software really fast.","7b9faa53b03b",[],"bc7ac67dfd67",[1427],"And so one of Brooks's assertions, of course, is the 1970s, but there's that the difference between a working software project and a software product can often be an order of magnitude.","8a94676c5fa1",[1456,1457,1458],[1450,1451],{"_212":1454,"_24":224,"_225":1455},{"_212":1452,"_24":224,"_225":1453},"9352f2e","00:03:51.360","0587358","00:03:41.920",{"_212":1464,"_24":233,"_234":1465,"_236":1466},{"_212":1462,"_24":233,"_234":1463,"_236":240},{"_212":1459,"_24":233,"_234":1460,"_236":1461},"539a146abb50",[1452],"and accidental complexity is everything that's introduced by your tools and programming languages and processes around it.","54d880f6d139",[],"87b9dcb8aa89",[1454],"And you often find yourself contending with these two different types of complexity: essential complexity, which is the stuff that you can't take away without making the product not work,","49d043da0985",[1476,1477,1478],[1470,1471],{"_212":1474,"_24":224,"_225":1475},{"_212":1472,"_24":224,"_225":1473},"1475084","00:03:35.360","35c9659","00:03:26.320",{"_212":1484,"_24":233,"_234":1485,"_236":1486},{"_212":1482,"_24":233,"_234":1483,"_236":240},{"_212":1479,"_24":233,"_234":1480,"_236":1481},"2111637c3429",[1472],"can result in a system that's very incoherent, without this conceptual integrity, where the parts don't make sense together.","8738eb1e319e",[],"038a40946d68",[1474],"A team that is too large or shares the responsibility of deciding, you know, what to build and how to design it amongst too many people, or not getting those people on the same page,","317a9e3a85f4",[1499,1500,1501,1502,1503],[1490,1491,1492],{"_212":1497,"_24":224,"_225":1498},{"_212":1495,"_24":224,"_225":1496},{"_212":1493,"_24":224,"_225":1494},"0b4dff2","00:03:22.040","3efc79e","00:03:12.939","78aada5","00:03:04.379",{"_212":1514,"_24":233,"_234":1515,"_236":1516},{"_212":1512,"_24":233,"_234":1513,"_236":240},{"_212":1509,"_24":233,"_234":1510,"_236":1511},{"_212":1507,"_24":233,"_234":1508,"_236":240},{"_212":1504,"_24":233,"_234":1505,"_236":1506},"c02f5eb45886",[1493],"that grows quadratically with the number of people involved.","5188cf5ef4bd",[],"92a65579e2e5",[1495],"So when you add people to a software project, paradoxically the team will sometimes become less productive because there's a lot more human coordination involved","c9d4af2cef9b",[],"264461917af2",[1497],"And in The Mythical Man-Month, to summarize it for you in about 60 seconds, the key idea is that coordination is part of the work.","a01e8d3797bc",[1520],[],{"_212":1521,"_24":233,"_234":1522,"_236":1523},"7641e7136061",[],"Coordination Is Part of the Work","dbc4b26420cd",[1533,1534,1535],[1527,1528],{"_212":1531,"_24":224,"_225":1532},{"_212":1529,"_24":224,"_225":1530},"7311743","00:02:55.840","944b922","00:02:48.040",{"_212":1541,"_24":233,"_234":1542,"_236":1543},{"_212":1539,"_24":233,"_234":1540,"_236":240},{"_212":1536,"_24":233,"_234":1537,"_236":1538},"d8e6f6ac32d9",[1529],"still impressive, but if you're a software company building software products, this is still very hard for a lot of reasons.","4d2bdc0b55d0",[],"d735c6c03f70",[1531],"So hitting the three hundred yard drive, impressive. Like asking Astra to one-shot a personal, you know, widget,","1a0a4b037939",[1550],[1547],{"_212":1548,"_24":224,"_225":1549},"02eeda2","00:02:36.739",{"_212":1551,"_24":233,"_234":1552,"_236":1553},"3a84ebf39442",[1548],"But to build software that is good, that is correct, that is maintainable by somebody other than you, that is good to use, that's in golf what they call the ground game, and that is still hard.","a9ba0ce5276e",[1563,1564,1565],[1557,1558],{"_212":1561,"_24":224,"_225":1562},{"_212":1559,"_24":224,"_225":1560},"8610700","00:02:30.819","36ba772","00:02:20.920",{"_212":1571,"_24":233,"_234":1572,"_236":1573},{"_212":1569,"_24":233,"_234":1570,"_236":240},{"_212":1566,"_24":233,"_234":1567,"_236":1568},"595ea0122f8a",[1559],"given, you know, non-golfers or non-coders the feeling of being able to hit a three hundred yard drive.","670807191c1c",[],"cbed60e094c3",[1561],"And so one of the best analogies that I like for what it's like to build software today is the golf metaphor, that agents have","4b7421abe0f3",[1583,1584,1585],[1577,1578],{"_212":1581,"_24":224,"_225":1582},{"_212":1579,"_24":224,"_225":1580},"56c7a1d","00:02:17.640","d2ef000","00:02:10.540",{"_212":1591,"_24":233,"_234":1592,"_236":1593},{"_212":1589,"_24":233,"_234":1590,"_236":240},{"_212":1586,"_24":233,"_234":1587,"_236":1588},"e53faddfceca",[1579],"That's a lot of clanker time.","900f9a55fb80",[],"192293f7a3dd",[1581],"But, you know, since the beginning of the year I've had on average two agents working twenty-four-seven the entire year.","458f6f877fe0",[1603,1604,1605],[1597,1598],{"_212":1601,"_24":224,"_225":1602},{"_212":1599,"_24":224,"_225":1600},"9d0b4a5","00:02:05.740","85adb9b","00:01:53.820",{"_212":1611,"_24":233,"_234":1612,"_236":1613},{"_212":1609,"_24":233,"_234":1610,"_236":240},{"_212":1606,"_24":233,"_234":1607,"_236":1608},"9067295fab2f",[1599],"and Dario, for all of the delightful subsidy, however long that will last.","4284266245eb",[],"267ba68e01b9",[1601],"I'm sure there's plenty of slop in there, but I've been consuming a tremendous amount of services, certainly not paying a quarter of a million dollars in tokens. Thank you, Sam","0724753996ea",[1629,1630,1631,1632,1633,1634,1635],[1617,1618,1619,1620],{"_212":1627,"_24":224,"_225":1628},{"_212":1625,"_24":224,"_225":1626},{"_212":1623,"_24":224,"_225":1624},{"_212":1621,"_24":224,"_225":1622},"7a6e308","00:01:45.580","c1a52ba","00:01:40.799","f2e4d31","00:01:38.560","a53b79e","00:01:34.960",{"_212":1651,"_24":233,"_234":1652,"_236":1653},{"_212":1649,"_24":233,"_234":1650,"_236":240},{"_212":1646,"_24":233,"_234":1647,"_236":1648},{"_212":1644,"_24":233,"_234":1645,"_236":240},{"_212":1641,"_24":233,"_234":1642,"_236":1643},{"_212":1639,"_24":233,"_234":1640,"_236":240},{"_212":1636,"_24":233,"_234":1637,"_236":1638},"24b8791054c8",[1621],"But yeah, I've been tremendously productive and could argue about whether this output is all good.","2f072efeb9dc",[],"07c304e9fc3e",[1623],"And I honestly think that I'm done writing code, and that's completely fine.","87b286b6a76e",[],"afb49c24f117",[1625],"I think it was sometime like last summer.","9621f2084205",[],"158f8f1bd152",[1627],"I had to look back when it was the last time I wrote code by hand.","fa49a0ca92f1",[1666,1667,1668,1669,1670],[1657,1658,1659],{"_212":1664,"_24":224,"_225":1665},{"_212":1662,"_24":224,"_225":1663},{"_212":1660,"_24":224,"_225":1661},"0cc36cd","00:01:32.440","f84f9c3","00:01:29.640","40bc48a","00:01:22.820",{"_212":1681,"_24":233,"_234":1682,"_236":1683},{"_212":1679,"_24":233,"_234":1680,"_236":240},{"_212":1676,"_24":233,"_234":1677,"_236":1678},{"_212":1674,"_24":233,"_234":1675,"_236":240},{"_212":1671,"_24":233,"_234":1672,"_236":1673},"aa84bf3c7cc2",[1660],"And I've definitely written no code this year.","184b23f40aca",[],"525caba1984d",[1662],"I am way down the rabbit hole.","59381c5317a4",[],"da030a54b8d3",[1664],"So why is that? And I'm not a skeptic by any means.","1b5959b2c6db",[1696,1697,1698,1699,1700],[1687,1688,1689],{"_212":1694,"_24":224,"_225":1695},{"_212":1692,"_24":224,"_225":1693},{"_212":1690,"_24":224,"_225":1691},"29990de","00:01:16.340","2f229bf","00:01:14.280","4531165","00:01:02.480",{"_212":1711,"_24":233,"_234":1712,"_236":1713},{"_212":1709,"_24":233,"_234":1710,"_236":240},{"_212":1706,"_24":233,"_234":1707,"_236":1708},{"_212":1704,"_24":233,"_234":1705,"_236":240},{"_212":1701,"_24":233,"_234":1702,"_236":1703},"e5eb1df86e37",[1690],"But if you look around, we're not swimming in amazing software built by normal people.","1ba772062241",[],"e7010773ac76",[1692],"It'll be great. There's nothing to worry about.","0679d1e1397d",[],"2e8b2d1902a1",[1694],"And there were people at the time who said to me, basically, this makes sense now, but in six months the models will get better, the harnesses will get better, and don't worry, normal people will be building amazing software.","5d1d9aa41303",[1723,1724,1725],[1717,1718],{"_212":1721,"_24":224,"_225":1722},{"_212":1719,"_24":224,"_225":1720},"182049b","00:00:51.920","2ebf61f","00:00:45.080",{"_212":1731,"_24":233,"_234":1732,"_236":1733},{"_212":1729,"_24":233,"_234":1730,"_236":240},{"_212":1726,"_24":233,"_234":1727,"_236":1728},"42336e966152",[1719],"And the argument that I made in the essay is that basically what's left is deciding what to build, how to structure it, and how to make it good.","d019106228f3",[],"57b6bda8085a",[1721],"So the agents have clearly taken away the mechanical part of building software, writing the code.","3c489fe1324b",[1749,1750,1751,1752,1753,1754,1755],[1737,1738,1739,1740],{"_212":1747,"_24":224,"_225":1748},{"_212":1745,"_24":224,"_225":1746},{"_212":1743,"_24":224,"_225":1744},{"_212":1741,"_24":224,"_225":1742},"a56c167","00:00:38.680","87f9ad0","00:00:35.600","fb31f14","00:00:32.000","3f03d02","00:00:22.720",{"_212":1771,"_24":233,"_234":1772,"_236":1773},{"_212":1769,"_24":233,"_234":1770,"_236":240},{"_212":1766,"_24":233,"_234":1767,"_236":1768},{"_212":1764,"_24":233,"_234":1765,"_236":240},{"_212":1761,"_24":233,"_234":1762,"_236":1763},{"_212":1759,"_24":233,"_234":1760,"_236":240},{"_212":1756,"_24":233,"_234":1757,"_236":1758},"febb28814611",[1741],"and talk about what's changed and what's remained the same in that time.","1ce670029cd7",[],"8eacbacf121b",[1743],"So I wanted to revisit the essay","d6d24551809d",[],"38f3f3e03783",[1745],"So it's been seven months, more than seven months, since I wrote this essay.","8c783856c99e",[],"21e371fd2e24",[1747],"It's from the nineteen seventies. And the idea was to look at what does it mean for software engineering when agents are writing all of the code.","2d363bcd783c",[1786,1787,1788,1789,1790],[1777,1778,1779],{"_212":1784,"_24":224,"_225":1785},{"_212":1782,"_24":224,"_225":1783},{"_212":1780,"_24":224,"_225":1781},"6402320","00:00:19.120","f03d50b","00:00:11.360","4909ed0","00:00:03.280",{"_212":1801,"_24":233,"_234":1802,"_236":1803},{"_212":1799,"_24":233,"_234":1800,"_236":240},{"_212":1796,"_24":233,"_234":1797,"_236":1798},{"_212":1794,"_24":233,"_234":1795,"_236":240},{"_212":1791,"_24":233,"_234":1792,"_236":1793},"723ce2371642",[1780],"And anyone doing engineering management should read the book.","514a07133b31",[],"34661033cd24",[1782],"And the idea was that there's a very famous book in software engineering, maybe the most famous book, by Fred Brooks called The Mythical Man-Month.","0760bbc53996",[],"a45d103e45cd",[1784],"So I wrote an essay in February called The Mythical Agent-Month, and so you could say to yourself that was a long time ago.","fd5bb8809e75",[1807],[],{"_212":1808,"_24":233,"_234":1809,"_236":1810},"e8df788faeaa",[],"Seven Months Later",{"_18":1826,"_20":1841,"_22":1828,"_24":1818,"_26":1829,"_1819":1842,"_1821":1843,"_14":1844},{"_18":1834,"_20":1835,"_22":1836,"_24":1818,"_26":1834,"_1819":1837,"_1821":1838,"_14":1839},{"_18":1826,"_20":1827,"_22":1828,"_24":1818,"_26":1829,"_1819":1830,"_1821":1831,"_14":1832},{"_18":1815,"_20":1816,"_22":1817,"_24":1818,"_26":1815,"_1819":1820,"_1821":1822,"_14":1823},"2023-07-17T23:23:49Z","eec559a9-10e7-4405-82e4-e0736495ec36","TfMkf5gVJ7UWv1vMi5Qxor","tag","description","Artificial intelligence, including AI-powered coding assistants built on large language models.","name","AI",{"_24":14,"_1824":1825},"current","ai","2026-10-23T23:10:26Z","8d2828f0-70fa-49f2-a7cf-4b648e941299","mG7vmCjaB1vEw3gqUIqrjf","2026-10-23T23:11:12Z","Coordinating many agents, goals, and roles.","Agent Orchestration",{"_24":14,"_1824":1833},"agent-orchestration","2025-04-15T17:19:22Z","80867d7d-9222-41f5-a465-915aa791bc97","YlxgmKTLxstUdop23VNy4f","AI agents are autonomous systems built to independently take action on business or technical tasks.","AI Agents",{"_24":14,"_1824":1840},"ai-agents","6625b640-07c5-4ac4-9324-bf13e52c5ed3","Agents that work toward goals over hours or days, and the infrastructure that keeps them on track.","Long-Horizon Agents",{"_24":14,"_1824":1845},"long-horizon-agents",{"_212":1847,"_24":1848,"_1849":-5,"_1850":1851},"spk0wesm","speaker","organization","person",{"_18":1852,"_20":1853,"_22":1854,"_1855":1856,"_24":1850,"_26":1857,"_1858":1859,"_1860":12,"_1821":1861,"_1849":1862,"_1863":12,"_1864":1865,"_1866":1867,"_68":1868,"_1869":12},"2026-03-01T20:38:52Z","1c82ff14-c3b0-4f57-858f-ed3328d727e8","IBgcBIPGoUVrOhZYuqwyIS","_system",{"_1945":1946},"2026-09-10T17:43:51Z","bio",[1883],"generate","Wes McKinney",{"_18":1877,"_20":1878,"_22":1879,"_24":1849,"_26":1880,"_1860":12,"_1821":1881,"_1882":12},"partner","photo",{"_24":1872,"_74":1873},"position","Cofounder",{"_24":68,"_1870":12,"_1871":12},"teamMember","nofollow","noindex","mainImage",{"_1874":1875,"_24":1876},"_ref","image-c483d6dccfb5903a5a2b786fb0c14893f9d09f77-450x450-jpg","reference","2026-09-10T17:40:01Z","466c3ec1-56fc-4758-9d87-f157819029ec","AvBrD6DdjyccEYuoF7qYYQ","2026-09-10T17:40:13Z","Kenn Software","portfolioCompany",{"_212":1884,"_24":214,"_215":1885,"_217":1886,"_219":220},"234c57c87a22",[1904,1905,1906,1907,1908,1909,1910,1911,1912,1913],[1887,1888,1889,1890,1891],{"_212":1902,"_24":1893,"_1894":1903},{"_212":1900,"_24":1893,"_1894":1901},{"_212":1898,"_24":1893,"_1894":1899},{"_212":1896,"_24":1893,"_1894":1897},{"_212":1892,"_24":1893,"_1894":1895},"e67849b690cc","link","href","https://posit.co/","ec07fc47e0b8","https://wesmckinney.com/book/","15e27da95938","https://arrow.apache.org/","2e54be5b0988","https://pandas.pydata.org/","5a34c3326d21","https://www.linkedin.com/in/wesmckinn/",{"_212":1942,"_24":233,"_234":1943,"_236":1861},{"_212":1939,"_24":233,"_234":1940,"_236":1941},{"_212":1936,"_24":233,"_234":1937,"_236":1938},{"_212":1933,"_24":233,"_234":1934,"_236":1935},{"_212":1930,"_24":233,"_234":1931,"_236":1932},{"_212":1927,"_24":233,"_234":1928,"_236":1929},{"_212":1923,"_24":233,"_234":1924,"_236":1925},{"_212":1920,"_24":233,"_234":1921,"_236":1922},{"_212":1917,"_24":233,"_234":1918,"_236":1919},{"_212":1914,"_24":233,"_234":1915,"_236":1916},"1554eecb3e4b",[],". ","6e9a3cce51d9",[1892],"Posit","0bbdd9a53463",[],", investor, and Principal Architect at ","4939de672f90",[1896,1926],"Python for Data Analysis","em","401909e79017",[],", a foundational standard for modern data infrastructure. He is a software engineer, author of ","4c935564ffa6",[1898],"Apache Arrow","16b4a5a11871",[]," and a co-creator of ","8cf8aa5d88f9",[1900],"Pandas","d7f13a35bac8",[]," is the creator of ","db7bce4890dc",[1902,1944],"strong","base",{"_100":1853,"_1947":1948},"rev","9nSj62b2W8O2pZi5wXkWtz","image",{"_1874":1951,"_24":1876},"image-332066ffd6da3f20904f5467f01cfd0ac2399f3e-1920x1080-jpg",{"_212":1953,"_24":214,"_215":1954,"_217":1955,"_219":220},"0a3733b36a56",[1956],[],{"_212":1957,"_24":233,"_234":1958,"_236":1959},"3d402b2a5cb6",[],"

The Mythical Agent-Month\n
- Seven Months Later
- Coordination Is Part of the Work
- Judgment, Taste, and Agency
- Taste and Decision Fatigue
- You Are the Agent
\n

",{"_212":1967,"_1874":1968,"_24":1876},{"_212":1965,"_1874":1966,"_24":1876},{"_212":1963,"_1874":1964,"_24":1876},"NT7XMfY0WQkS","39c370db-08bf-4de5-b2c2-7b579fe65e59","bYOZb2fH1KCp","67355535-cdb5-472f-b416-5b48b1294f09","s4U4876ytCUN","4cf0dd20-6e50-4e54-91f2-18e60d34af37","items",[4429,4430,4431],"recent",[1973,1974,1975],{"_1976":3811,"_96":3812,"_1949":-5,"_1980":3813,"_51":3814,"_14":3815,"_1984":1985,"_63":3816,"_1987":1988},{"_1976":2592,"_96":2593,"_1949":-5,"_1980":2594,"_51":2595,"_14":2596,"_1984":1985,"_63":2597,"_1987":1988},{"_1976":1977,"_96":1978,"_1949":1979,"_1980":1981,"_51":1982,"_14":1983,"_1984":1985,"_63":1986,"_1987":1988},"content",[1989,1990,1991,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,2002,2003,2004,2005,2006,2007,2008,2009,2010,2011,2012,2013,2014,2015,2016,2017,2018,2019,2020,2021,2022,2023,2024,2025,2026,2027,2028,2029,2030,2031,2032,2033],17,"https://cdn.sanity.io/images/50q6fr1p/production/28801f05c737cc57c3c04d467a0a54838fc70ad0-1200x630.png","objectID","9d8377ac-25e6-47b7-8e40-22ee9a7cb43e","2026-08-27T16:38:00.000Z","understanding-code-by-treating-it-as-a-graph","templateType","default","Is the Key to Understanding Code Treating it as a Graph?","typeName","blogArticle",{"_212":2585,"_24":214,"_215":2586,"_217":2587,"_219":598},{"_212":2556,"_24":214,"_215":2557,"_217":2558,"_219":220},{"_212":2527,"_24":214,"_215":2528,"_217":2529,"_219":220},{"_212":2524,"_24":1949,"_74":2525},{"_212":2511,"_24":214,"_215":2512,"_217":2513,"_219":220},{"_212":2504,"_24":214,"_215":2505,"_217":2506,"_219":598},{"_212":2486,"_24":214,"_215":2487,"_217":2488,"_219":220},{"_212":2468,"_24":214,"_215":2469,"_217":2470,"_219":220},{"_212":2461,"_24":214,"_215":2462,"_217":2463,"_219":220},{"_212":2454,"_24":214,"_215":2455,"_217":2456,"_219":220},{"_212":2447,"_24":214,"_215":2448,"_217":2449,"_219":598},{"_212":2418,"_24":214,"_215":2419,"_217":2420,"_219":220},{"_212":2411,"_24":214,"_215":2412,"_217":2413,"_219":220},{"_212":2404,"_24":214,"_215":2405,"_217":2406,"_219":220},{"_212":2397,"_24":214,"_215":2398,"_217":2399,"_219":598},{"_212":2390,"_24":214,"_215":2391,"_217":2392,"_219":220},{"_212":2372,"_24":214,"_215":2373,"_217":2374,"_219":220},{"_212":2354,"_24":214,"_215":2355,"_217":2356,"_219":220},{"_212":2347,"_24":214,"_215":2348,"_217":2349,"_219":598},{"_212":2340,"_24":214,"_215":2341,"_217":2342,"_219":220},{"_212":2314,"_24":214,"_215":2315,"_217":2316,"_219":220},{"_212":2296,"_24":214,"_215":2297,"_217":2298,"_219":220},{"_212":2281,"_24":214,"_215":2282,"_217":2283,"_219":220},{"_212":2274,"_24":214,"_215":2275,"_217":2276,"_219":220},{"_212":2267,"_24":214,"_215":2268,"_217":2269,"_219":220},{"_212":2260,"_24":214,"_215":2261,"_217":2262,"_219":598},{"_212":2253,"_24":214,"_215":2254,"_217":2255,"_219":220},{"_212":2246,"_24":214,"_215":2247,"_217":2248,"_219":220},{"_212":2239,"_24":214,"_215":2240,"_217":2241,"_219":220},{"_212":2210,"_24":214,"_215":2211,"_217":2212,"_219":220},{"_212":2203,"_24":214,"_215":2204,"_217":2205,"_219":598},{"_212":2196,"_24":214,"_215":2197,"_217":2198,"_219":220},{"_212":2189,"_24":214,"_215":2190,"_217":2191,"_219":220},{"_212":2182,"_24":214,"_215":2183,"_217":2184,"_219":220},{"_212":2175,"_24":214,"_215":2176,"_217":2177,"_219":220},{"_212":2168,"_24":214,"_215":2169,"_217":2170,"_219":598},{"_212":2161,"_24":214,"_215":2162,"_217":2163,"_219":220},{"_212":2121,"_24":214,"_215":2122,"_217":2123,"_219":220},{"_212":2092,"_24":214,"_215":2093,"_217":2094,"_219":220},{"_212":2085,"_24":214,"_215":2086,"_217":2087,"_219":598},{"_212":2078,"_24":214,"_215":2079,"_217":2080,"_219":220},{"_212":2063,"_24":214,"_215":2064,"_217":2065,"_219":220},{"_212":2056,"_24":214,"_215":2057,"_217":2058,"_219":220},{"_212":2041,"_24":214,"_215":2042,"_217":2043,"_219":220},{"_212":2034,"_24":214,"_215":2035,"_217":2036,"_219":220},"8eaf5dda166e",[2037],[],{"_212":2038,"_24":233,"_234":2039,"_236":2040},"7fccb285f275",[],"“So I think orchestration will be important for quite some time, because AI is still not at the level of fully, autonomously orchestrating large-scale systems that will work smoothly and without any problems. We are not there yet.”","628421de9da8",[2044,2045,2046],[],{"_212":2053,"_24":233,"_234":2054,"_236":2055},{"_212":2050,"_24":233,"_234":2051,"_236":2052},{"_212":2047,"_24":233,"_234":2048,"_236":2049},"ad91e713dbfb",[],", but also understanding what caching is and why it’s important. You’re going to need to apply that understanding when you think your system actually needs caching!”","d4b8c4a82cf1",[1926],"what Redis is","0cc52c592750",[],"“It’s about having this type of ‘systems thinking,’ an understanding of system design architecture and the underlying principles of the technologies. Not just knowing ","dc75d7ad8097",[2059],[],{"_212":2060,"_24":233,"_234":2061,"_236":2062},"c6a9980c57d2",[],"“You need to understand how the distributed systems work, and the code-graph-rag project fits really well here to help you ask architectural questions and locate potential bottlenecks, like dead code or other opportunities for optimization.”","afb3c879f969",[2066,2067,2068],[],{"_212":2075,"_24":233,"_234":2076,"_236":2077},{"_212":2072,"_24":233,"_234":2073,"_236":2074},{"_212":2069,"_24":233,"_234":2070,"_236":2071},"56e7f92d1742",[]," to use it, and what the pros and cons are. That will help you to have this overarching, holistic view of what it's like to build a highly scalable software engineering system.”","7f5989d240bd",[1926],"when","49750bd53c8c",[],"“For example, in order to deploy applications at scale, you might need a Kubernetes cluster. You shouldn’t have to be a highly skilled DevOps engineer to know that as AI can help you there, but you would need to know ","ab8bfa185ba6",[2081],[],{"_212":2082,"_24":233,"_234":2083,"_236":2084},"067bce70e73c",[],"From his experience building the project, Avagyan advises anyone working in a technical capacity to learn how to orchestrate. “Software engineering is changing faster than people think. The ability to orchestrate things will be important no matter what. By that, I mean knowing many different technologies: Their strengths and weaknesses, the scalability of each thing, and when to use what.”","1d6c6865362e",[2088],[],{"_212":2089,"_24":233,"_234":2090,"_236":2091},"2586e6d9e8b9",[],"The Future of Codebases: Built and Managed by AI?","6869a525750a",[2101,2102,2103,2104,2105],[2095,2096],{"_212":2099,"_24":1893,"_1894":2100},{"_212":2097,"_24":1893,"_1894":2098},"41760ee29615","https://www.youtube.com/watch?v=O2jUTq6nCEY&t=2s","48182a105088","https://blog.dailydoseofds.com/",{"_212":2118,"_24":233,"_234":2119,"_236":2120},{"_212":2115,"_24":233,"_234":2116,"_236":2117},{"_212":2112,"_24":233,"_234":2113,"_236":2114},{"_212":2109,"_24":233,"_234":2110,"_236":2111},{"_212":2106,"_24":233,"_234":2107,"_236":2108},"dd3d4a61f74f",[],"), but eventually, the project started trending again on GitHub organically. I guess it might have to do with the ‘right solution at the right time’ phenomenon.”","6f0d56991c70",[2097],"with Memgraph","3954a9df1f48",[]," when I was working on my own startup back in 2023. I got coverage for my product in this popular newsletter, which might have been the trigger that helped it ‘go viral’ on GitHub the first time. That said, I have also tried many other channels like Hacker News, LinkedIn and X posts, giving talks (including this session ","6424a1b96d65",[2099],"Daily Dose of Data Science","30637c109bb6",[],"“Finally, I’d recommend finding a distribution channel for your work. Even the greatest product on earth won’t necessarily be noticed if you just put it on GitHub and leave it there. You need to promote it. I was fortunate enough to have worked with the founder of ","81d7413fc3b2",[2133,2134,2135,2136,2137,2138,2139],[2124,2125,2126],{"_212":2131,"_24":1893,"_1894":2132},{"_212":2129,"_24":1893,"_1894":2130},{"_212":2127,"_24":1893,"_1894":2128},"2f238ed440f7","https://zed.dev/","3f971d287b17","https://neovim.io/","f040745fb71d","https://github.com/vitali87/croft",{"_212":2158,"_24":233,"_234":2159,"_236":2160},{"_212":2155,"_24":233,"_234":2156,"_236":2157},{"_212":2152,"_24":233,"_234":2153,"_236":2154},{"_212":2149,"_24":233,"_234":2150,"_236":2151},{"_212":2146,"_24":233,"_234":2147,"_236":2148},{"_212":2143,"_24":233,"_234":2144,"_236":2145},{"_212":2140,"_24":233,"_234":2141,"_236":2142},"1bb94c846e83",[]," and other alternatives. I found myself thinking there should be a better way, so I built a VS Code clone in the terminal with Rust to make it super-fast. I’ve been using it every day myself and I would recommend others try it out.”","93a91dcf4857",[2127],"Zed","e76bcfb5a8f3",[],", ","94d7ac6fa32a",[2129],"Neovim","630e70cbbe50",[],", as an example. “I’m a VS Code user, but I enjoy terminal as well, so I found myself going back and forth between VS Code, ","9f8ed6fdab01",[2131],"Croft","321ac6bc0e40",[],"“Second, I would recommend building things from the perspective of thinking about what you do every day and how you can accelerate what you do and be more performant with your tools.” The creator notes another open-source project of his, ","8c4981547047",[2164],[],{"_212":2165,"_24":233,"_234":2166,"_236":2167},"58335d9df518",[],"When asked what his most important learnings from the project are, Avagyan offers a few suggestions. “One, which I think is most important, is to just build something for yourself that you would like to have. That's pretty much how this project was born. Because I was dealing with [the challenges of managing a] monorepo, and at first, I just wanted to visualize it and understand its complexity. I looked for solutions to see if there was a good product for me. I didn't find one, so I decided I would build it myself, and after releasing it publicly, it turned to be useful for other people.”","05e8e1c9fe0a",[2171],[],{"_212":2172,"_24":233,"_234":2173,"_236":2174},"76872200c049",[],"Learnings from Building","526da0bc67f6",[2178],[],{"_212":2179,"_24":233,"_234":2180,"_236":2181},"b5c434461aac",[],"The creator suggests that while he spent the better part of a year building a system that could dynamically map and remap graph representations of codebases without significant latency, not every org is on the same page. “Even doing something like this with AI agents will take some time, and will cost internal resources. And people don’t always have patience for these things and may want to see a one-shot solution instead. But I think the approach will get more momentum and I think many companies will adopt it eventually.”","f2f91f55c9c2",[2185],[],{"_212":2186,"_24":233,"_234":2187,"_236":2188},"5556895009c6",[],"“You can get really far with one language, maybe with a toy example using a tool like Tree-sitter. But jobs that require you to unify graphs across different languages, and build usable and accurate relationships over top of them, and making sure it can work at scale, that’s different. For example, a monorepo is a living creature that changes every minute: People commit and delete code and so on. So your graph needs to keep up with that. It needs to recalculate [those relationships] with every change, every drop, every addition. The project has a graph updater designed specifically to make this real-time syncing possible.”","a89686ddcbd7",[2192],[],{"_212":2193,"_24":233,"_234":2194,"_236":2195},"eb00873aac4a",[],"Avagyan notes that those companies may be hitting a stumbling block. “I think one problem companies may be having is accuracy.”","7cbaec07cf9b",[2199],[],{"_212":2200,"_24":233,"_234":2201,"_236":2202},"52c6bc34d2b2",[],"The creator suggests that graphing codebases is a useful approach for any organization, but the timing might not be quite right for large-scale enterprises that are slow to embrace change. “I believe many companies, including major AI vendors like Cursor or Anthropic, have been experimenting with graph-based approaches, if not for codebase mapping, then for use cases like memory management.”","5a3ea8acd83b",[2206],[],{"_212":2207,"_24":233,"_234":2208,"_236":2209},"1cd77c6dd0f6",[],"Why Aren’t Enterprises Already Graphing Their Codebases?","b003dd295bbf",[2219,2220,2221,2222,2223],[2213,2214],{"_212":2217,"_24":1893,"_1894":2218},{"_212":2215,"_24":1893,"_1894":2216},"eb5b74dd23fd","https://flutter.dev/","fc884ca87ef9","https://dart.dev/overview",{"_212":2236,"_24":233,"_234":2237,"_236":2238},{"_212":2233,"_24":233,"_234":2234,"_236":2235},{"_212":2230,"_24":233,"_234":2231,"_236":2232},{"_212":2227,"_24":233,"_234":2228,"_236":2229},{"_212":2224,"_24":233,"_234":2225,"_236":2226},"b630193673be",[]," widget structure.”","ab660c7a1ffd",[2215],"Flutter","cee0616ee44a",[]," is already supported and includes ","fc85ec3d18b6",[2217],"Dart","2a8e4e8ba899",[],"Avagyan notes growing interest among teams managing heavyweight ERP systems with some flavor of Python, as well as ongoing requests around app development. “","e96ed7f2d62e",[2242],[],{"_212":2243,"_24":233,"_234":2244,"_236":2245},"e4b2c56f631d",[],"So you can see why companies are adopting the project as part of their onboarding tools. You could use it as an MCP server in Claude Code or any other harness or just use Cypher queries to search your codebase before even starting any agentic exploration. This approach is quick and gives you a really good grasp of a codebase, without getting sidetracked to other areas that may not be as important.”","a9bdfd671d83",[2249],[],{"_212":2250,"_24":233,"_234":2251,"_236":2252},"b0569066d73c",[],"“Since we recently added FLOWS_TO, READS_FROM and WRITES_TO relationships in a graph, it’s now easy to analyze the relationships between microservices in your codebase as well. (I’ve actually been approached by security companies about using the project for security detection use cases across large codebases, which is a natural fit as tracing how data moves through a system is exactly what those relationships capture.)”","34664692f09c",[2256],[],{"_212":2257,"_24":233,"_234":2258,"_236":2259},"2b1eaf954306",[],"The creator calls out onboarding new engineers as a key use case for the project. “People tell me that using the project has significantly reduced onboarding times because it can gather information about different aspects of a monorepo codebase and give that directly to new junior engineers.”","06fc03e6d208",[2263],[],{"_212":2264,"_24":233,"_234":2265,"_236":2266},"0743e3b412d6",[],"Real-World Use Cases for Graphing a Codebase","6a4ddfd50139",[2270],[],{"_212":2271,"_24":233,"_234":2272,"_236":2273},"847581b898f9",[],"“To be honest, the goal of the project was never cost reduction, but I think it’s a byproduct, since you get the accuracy of a graph and the accuracy of an MCP and agent to invoke the correct tools to query the graph.","2cfb4c4a9a81",[2277],[],{"_212":2278,"_24":233,"_234":2279,"_236":2280},"b8c18965ccb1",[],"“Let’s say the project fails in one Cypher query? Then it can just repeat it, and with that query, you can gain lots of context up front, very fast. And that failure is less costly compared to if AI were to go and perform its own search, realize it was searching in the wrong place, then go to another place...you can imagine how things would accumulate really fast.”","824172ebddac",[2284,2285,2286],[],{"_212":2293,"_24":233,"_234":2294,"_236":2295},{"_212":2290,"_24":233,"_234":2291,"_236":2292},{"_212":2287,"_24":233,"_234":2288,"_236":2289},"e051511f75df",[]," say that I’ve observed the project to be very fast, and it seems more fault-tolerant in a way.”","36ec44554375",[1926],"can","7cdd05898655",[],"Avagyan notes that some users come to the project looking for performance and cost gains as well. “Some people claim that the project is as much as 65% more cost-efficient than traditional agentic coding and search, and I can’t verify those numbers myself. I ","623b0301e0bf",[2302,2303,2304],[2299],{"_212":2300,"_24":1893,"_1894":2301},"c1ed526f5769","https://memgraph.com/",{"_212":2311,"_24":233,"_234":2312,"_236":2313},{"_212":2308,"_24":233,"_234":2309,"_236":2310},{"_212":2305,"_24":233,"_234":2306,"_236":2307},"da89921f17d9",[]," as a GraphDB and it's really fast, it runs in memory and you can deploy using Docker containers, and that would be enough to deploy locally, and the project has the building blocks for bigger deployments to the cloud. So far, it seems to scale well (people have even graphed the Linux kernel) and I haven’t seen any real degradation in speed when it comes to querying graphs.”","34da333815d3",[2300],"Memgraph","c9fdbc40e4d6",[],"“Otherwise, I don't think storage is a problem. The project uses ","d8236069c0ba",[2320,2321,2322,2323,2324],[2317],{"_212":2318,"_24":1893,"_1894":2319},"ec14de8b6db9","https://en.wikipedia.org/wiki/Dead-code_elimination",{"_212":2337,"_24":233,"_234":2338,"_236":2339},{"_212":2334,"_24":233,"_234":2335,"_236":2336},{"_212":2331,"_24":233,"_234":2332,"_236":2333},{"_212":2328,"_24":233,"_234":2329,"_236":2330},{"_212":2325,"_24":233,"_234":2326,"_236":2327},"2dd3e29f6f5e",[]," command that will check the reachability of all your code. If some are not reachable, they’re likely candidates [for being dead code and potentially getting deleted].”","cba21817906a",[1926],"code graph","28ed3f1a3a15",[],". If you have vibe coded your repo, or it’s otherwise in a state that you don't know exactly what's happening because there are so many abstractions or code duplications, or that code isn’t reachable. For example, there is a ","0f95656bd1a7",[2318],"dead code detection","37e4c8419854",[],"“I would say that a practical use of code-graph-rag is ","2e27e83e6ea0",[2343],[],{"_212":2344,"_24":233,"_234":2345,"_236":2346},"b4e8f24bb41b",[],"The creator suggests that the project is continuing to improve on traditional challenges in graph-based projects, which have historically not always been particularly responsive. “I would say storage has never been a problem. With this project, I spent significant time optimizing the build process of going from codebase to graph. (I actually optimised the build time by more than 5x with a separate repo I built that optimizes user-defined codebase segments against specific metrics like speed within accepted memory constraints. The agent then \"autosearches\" the best code among competing candidates while running the build against the ground-truthed evaluation cases which you or your agents have collected or constructed yourself.)”","e71e561bda30",[2350],[],{"_212":2351,"_24":233,"_234":2352,"_236":2353},"8bfe3199c316",[],"Bottlenecks and Resource Management","9b53b53a2ef7",[2360,2361,2362],[2357],{"_212":2358,"_24":1893,"_1894":2359},"cb676615c060","https://simonwillison.net/guides/agentic-engineering-patterns/red-green-tdd/",{"_212":2369,"_24":233,"_234":2370,"_236":2371},{"_212":2366,"_24":233,"_234":2367,"_236":2368},{"_212":2363,"_24":233,"_234":2364,"_236":2365},"a3e38bb4ef42",[],". Over time, the combination of hand-built code and agent-powered testing and development has proven to be effective. “At the beginning, I spent so much time on architecting [the project], setting the right relationships, that sort of thing, to ensure the foundation was ‘right.’ But when the foundation is ‘right,’ AI is great at doing the rest.”","a90a0f3ba6d2",[2358],"red/green test-driven development","84140f1e533e",[],"“And from there, we run thousands of test cases to ensure there are no regressions every time a new functionality is added. Every new functionality or bugfix is accompanied by a handful of new tests.” The creator suggests the project continuously runs more than 7,000 unit/integration tests using ","4ed1d30f3dc2",[2378,2379,2380],[2375],{"_212":2376,"_24":1893,"_1894":2377},"1b1e54e0707e","https://en.wikipedia.org/wiki/Polymorphism_(computer_science)",{"_212":2387,"_24":233,"_234":2388,"_236":2389},{"_212":2384,"_24":233,"_234":2385,"_236":2386},{"_212":2381,"_24":233,"_234":2382,"_236":2383},"9b109a84fb06",[]," works.” The creator notes that tracking all the nuances isn’t necessarily covered by existing deterministic open-source tools either. What is even more difficult is the runtime behavior of the code. That's why recent efforts have been placed to trace call relationships apparent at runtime and merge them into the same graph with statically-derived ones.”","e56a902ff095",[2376],"polymorphism","9a4ee9a258b2",[],"“Obviously, there are so many considerations when you go into specifics, like finding call relationships. Each language has its own specific intricacies, like how ","0815be5d9aee",[2393],[],{"_212":2394,"_24":233,"_234":2395,"_236":2396},"6ea22a083826",[],"As he sought to ensure the project yielded reliable graph representations, Avagyan found that asking LLMs to one-shot graph representations didn’t quite work. “[The LLMs] hallucinated some relationship edges and nodes. So, the project uses Tree-sitter parsers for different languages, which give you the basic relationships you can extract from AST as something of a unified way of using any type of language.”","6154d54ed4e6",[2400],[],{"_212":2401,"_24":233,"_234":2402,"_236":2403},"1641bec46e7a",[],"Making Code Graphs Deterministic with AST","a36ee5a77fd2",[2407],[],{"_212":2408,"_24":233,"_234":2409,"_236":2410},"b5e47a104baa",[],"“Obviously, this has lots of benefits. If you ask a question about a specific function or a service with semantic search, it looks at the vector similarity and brings you some functions, some of which may have relevant names, but may not be relevant to your actual query. But using graphs, if you can translate your intent into Cypher (with the help of AI), you can basically pinpoint which function you are looking for.” The creator clarifies that the project also supports semantic search as an add-on to support things like code comments, and that in practice, graph parsing and semantic search tend to be complementary.","4b92136c2a79",[2414],[],{"_212":2415,"_24":233,"_234":2416,"_236":2417},"916fbfa96bd2",[],"Avagyan’s approach was to unify the variety within a single graph ontology that worked across multiple languages, from which he designed the project to let users map relationships within a monorepo, even those that contain multiple languages (the project currently supports more than a dozen). With such precise relationships properly mapped, users can use Cypher or another query language (or some sort of tool to map natural language to a query language) to query any part of a codebase as a graph, and compare against semantic search.","755c01bddb6f",[2427,2428,2429,2430,2431],[2421,2422],{"_212":2425,"_24":1893,"_1894":2426},{"_212":2423,"_24":1893,"_1894":2424},"249727404c78","https://en.wikipedia.org/wiki/Abstract_syntax_tree","d7c41e85c164","https://tree-sitter.github.io/tree-sitter/",{"_212":2444,"_24":233,"_234":2445,"_236":2446},{"_212":2441,"_24":233,"_234":2442,"_236":2443},{"_212":2438,"_24":233,"_234":2439,"_236":2440},{"_212":2435,"_24":233,"_234":2436,"_236":2437},{"_212":2432,"_24":233,"_234":2433,"_236":2434},"916d54525867",[]," and gives you all the syntactic relationships, it's not still enough because there’s so much variety between languages and frameworks.”","ecb02d6d29a7",[2423],"abstract syntax tree","2d414b50c68b",[],", which parses the ","c4f79c67c17a",[2425],"Tree-sitter","8cfad1aaed2e",[],"The creator clarifies that precisely mapping codebases to graphs requires a correct build for each framework and programming language. “Even with an underlying technology like ","cc2e0c67c8ad",[2450],[],{"_212":2451,"_24":233,"_234":2452,"_236":2453},"a7a96359e940",[],"Accounting for Variation Across Languages, Frameworks","55289533b05a",[2457],[],{"_212":2458,"_24":233,"_234":2459,"_236":2460},"494b6d9eb54a",[],"The creator suggests that the project specifically addresses the dichotomy between graph search and traditional text search/grep. “The first part of the project is how you go from code to graph. The second is an assistant that has a context of a graph, but it also has all the tools that typical coding agents or harnesses have. I also use it alongside my other coding harnesses. For some problems, it can find solutions faster and more accurately.”","8edb1e48b5a2",[2464],[],{"_212":2465,"_24":233,"_234":2466,"_236":2467},"b5cb7e6813d4",[],"The creator clarifies, “that doesn't mean you should use [text] or [graphs] exclusively. They can be used in conjunction with each other. You can have grep, other types of search, and graph search together. Then, given that the AI has both contexts, it will decide which way to go.”","6bfeff21d7b3",[2474,2475,2476],[2471],{"_212":2472,"_24":1893,"_1894":2473},"40d5a9d1fab0","https://en.wikipedia.org/wiki/Cypher_(query_language)",{"_212":2483,"_24":233,"_234":2484,"_236":2485},{"_212":2480,"_24":233,"_234":2481,"_236":2482},{"_212":2477,"_24":233,"_234":2478,"_236":2479},"850209620fae",[]," query that will calculate it for you in milliseconds.” The creator suggests that treating codebases as graphs in this way can conserve both tokens and time.","fd7434afa3fb",[2472],"Cypher","2814a2a422f6",[],"The creator gives the example of asking an agent to find the longest nested function call in your codebase. Today’s agents will search through the imports and invocations and do some counting, a task that LLMs aren’t great at. “But if you already have a graph that represents all core relationships between methods and functions, you or an AI agent can just write a ","c357da4c0347",[2492,2493,2494],[2489],{"_212":2490,"_24":1893,"_1894":2491},"67f4fbd2fe9a","https://en.wikipedia.org/wiki/Grep",{"_212":2501,"_24":233,"_234":2502,"_236":2503},{"_212":2498,"_24":233,"_234":2499,"_236":2500},{"_212":2495,"_24":233,"_234":2496,"_236":2497},"7de49dd818b7",[]," or possibly some kind of agentic search. As you might imagine, that can become inefficient, especially considering some of the questions you might typically have about a codebase.”","683e39e682aa",[2490],"grep","b42b4a2c8abb",[],"Avagyan notes that when you hand a codebase to one of today’s coding assistants, it’ll often treat it as a text artifact. “Then, it traverses the code base with a ","2fc0e81b9ef7",[2507],[],{"_212":2508,"_24":233,"_234":2509,"_236":2510},"4cb4e6406d24",[],"Viewing Code as a Retrievable Graph","b80310699127",[2517,2518],[2514],{"_212":2515,"_24":1893,"_1894":2516},"5b72c578e5d1","https://www.youtube.com/watch?v=O2jUTq6nCEY",{"_212":2521,"_24":233,"_234":2522,"_236":2523},{"_212":2519,"_24":233,"_234":2520,"_236":2310},"5ef047d4ccf9",[2515,1926],"a0931bc06a15",[1926],"Vitali Avagyan discusses using GraphRAG to parse repos with Memgraph. Image courtesy ","c9d37b6922a5",{"_1874":2526,"_24":1876},"image-bc24d7b1e62740bb7448a1a80a9dfd59940c68d6-3840x1100-jpg","9b432548c002",[2536,2537,2538,2539,2540],[2530,2531],{"_212":2534,"_24":1893,"_1894":2535},{"_212":2532,"_24":1893,"_1894":2533},"b6ce27c00ce8","https://github.com/vitali87/code-graph-rag","a50030cc5b04","https://www.linkedin.com/in/vitali-avagyan-a1566234/",{"_212":2553,"_24":233,"_234":2554,"_236":2555},{"_212":2550,"_24":233,"_234":2551,"_236":2552},{"_212":2547,"_24":233,"_234":2548,"_236":2549},{"_212":2544,"_24":233,"_234":2545,"_236":2546},{"_212":2541,"_24":233,"_234":2542,"_236":2543},"c24ce7265e1d",[],".","c66f9ef7a776",[2532],"code-graph-rag","a0f94686e2c4",[]," asserts that the key to understanding code, particularly code across different languages, is to stop thinking of it as lines of plain text, and to treat it as a relational graph, which is the basis of his open-source project ","7aa3b54f64f6",[2534],"Vitali Avagyan","9c50f1246f14",[],"As generating code becomes free, understanding what any of it does becomes much more costly, and updating it becomes significantly more complex. Developer ","4b5755f9b8bc",[2565,2566,2567,2568,2569],[2559,2560],{"_212":2563,"_24":1893,"_1894":2564},{"_212":2561,"_24":1893,"_1894":2562},"8042bdcd04fd","https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/cloud-next-2026-sundar-pichai/","afe823ab102f","https://techcrunch.com/2025/04/29/microsoft-ceo-says-up-to-30-of-the-companys-code-was-written-by-ai/",{"_212":2582,"_24":233,"_234":2583,"_236":2584},{"_212":2579,"_24":233,"_234":2580,"_236":2581},{"_212":2576,"_24":233,"_234":2577,"_236":2578},{"_212":2573,"_24":233,"_234":2574,"_236":2575},{"_212":2570,"_24":233,"_234":2571,"_236":2572},"70dd8d1c5fc1",[]," of their new code is AI-generated, which does nothing to help the ongoing challenges of first-time onboarding for codebases and the updating of existing codebases for newly-hired engineers and growing teams (as well as for agentic coding assistants).","a7022507758b",[2561],"75%","47cf0e8a331c",[]," to ","d1aebf2378c0",[2563],"30%","adc4d52ad4cd",[],"Major engineering orgs claim that as much as ","f09e9612ef15",[2588],[],{"_212":2589,"_24":233,"_234":2590,"_236":2591},"85de00f41251",[],"Why Code-as-Text Doesn’t Work for Genuine Understanding",[2598,2599,2600,2601,2602,2603,2604,2605,2606,2607,2608,2609,2610,2611,2612,2613,2614,2615,2616,2617,2618,2619,2620,2621,2622,2623,2624,2625,2626,2627,2628,2629,2630,2631,2632,2633,2634,2635,2636,2637,2638,2639,2640,2641,2642,2643,2644,2645,2646,2647,2648,2649,2650,2651,2652,2653,2654,2655,2656,2657,2658,2659,2660,2661,2662,2663,2664,2665,2666,2667,2668,2669,2670,2671,2672,2673,2674,2675,2676,2677,2678,2679,2680,2681,2682,2683,2684,2685],20,"7f2db4aa-8e93-45db-a467-a326acd92cfb","2026-10-02T19:09:00.000Z","long-horizon-agents-part-2-memory-is-more-than-recall","Long-Horizon Agents, Part 2: Memory Is More than Recall",{"_212":3804,"_24":214,"_215":3805,"_217":3806,"_219":598},{"_212":3798,"_24":214,"_215":3799,"_217":3800,"_219":220},{"_212":3780,"_24":214,"_215":3781,"_217":3782,"_219":220},{"_212":3773,"_24":214,"_215":3774,"_217":3775,"_219":220},{"_212":3766,"_24":214,"_215":3767,"_2702":2703,"_2704":2705,"_217":3768,"_219":220},{"_212":3759,"_24":214,"_215":3760,"_2702":2703,"_2704":2705,"_217":3761,"_219":220},{"_212":3752,"_24":214,"_215":3753,"_2702":2703,"_2704":2705,"_217":3754,"_219":220},{"_212":3737,"_24":214,"_215":3738,"_217":3739,"_219":220},{"_212":3730,"_24":214,"_215":3731,"_217":3732,"_219":598},{"_212":3723,"_24":214,"_215":3724,"_217":3725,"_219":220},{"_212":3716,"_24":214,"_215":3717,"_217":3718,"_219":220},{"_212":3709,"_24":214,"_215":3710,"_217":3711,"_219":220},{"_212":3702,"_24":214,"_215":3703,"_217":3704,"_219":377},{"_212":3696,"_24":214,"_215":3697,"_217":3698,"_219":377},{"_212":3689,"_24":214,"_215":3690,"_217":3691,"_219":377},{"_212":3682,"_24":214,"_215":3683,"_217":3684,"_219":377},{"_212":3675,"_24":214,"_215":3676,"_217":3677,"_219":377},{"_212":3668,"_24":214,"_215":3669,"_217":3670,"_219":377},{"_212":3661,"_24":214,"_215":3662,"_217":3663,"_219":377},{"_212":3655,"_24":214,"_215":3656,"_217":3657,"_219":377},{"_212":3648,"_24":214,"_215":3649,"_217":3650,"_219":377},{"_212":3641,"_24":214,"_215":3642,"_217":3643,"_219":377},{"_212":3634,"_24":214,"_215":3635,"_217":3636,"_219":377},{"_212":3627,"_24":214,"_215":3628,"_217":3629,"_219":377},{"_212":3620,"_24":214,"_215":3621,"_217":3622,"_219":377},{"_212":3613,"_24":214,"_215":3614,"_217":3615,"_219":377},{"_212":3606,"_24":214,"_215":3607,"_217":3608,"_219":377},{"_212":3599,"_24":214,"_215":3600,"_217":3601,"_219":220},{"_212":3592,"_24":214,"_215":3593,"_217":3594,"_219":220},{"_212":3575,"_24":214,"_215":3576,"_217":3577,"_219":220},{"_212":3572,"_24":1949,"_74":3573},{"_212":3554,"_24":214,"_215":3555,"_217":3556,"_219":220},{"_212":3547,"_24":214,"_215":3548,"_217":3549,"_219":220},{"_212":3540,"_24":214,"_215":3541,"_217":3542,"_219":598},{"_212":3533,"_24":214,"_215":3534,"_217":3535,"_219":220},{"_212":3511,"_24":214,"_215":3512,"_217":3513,"_219":220},{"_212":3481,"_24":214,"_215":3482,"_217":3483,"_219":220},{"_212":3422,"_24":214,"_215":3423,"_217":3424,"_219":220},{"_212":3404,"_24":214,"_215":3405,"_217":3406,"_219":220},{"_212":3397,"_24":214,"_215":3398,"_217":3399,"_219":598},{"_212":3390,"_24":214,"_215":3391,"_217":3392,"_219":220},{"_212":3341,"_24":214,"_215":3342,"_217":3343,"_219":220},{"_212":3334,"_24":214,"_215":3335,"_217":3336,"_219":220},{"_212":3307,"_24":214,"_215":3308,"_217":3309,"_219":220},{"_212":3285,"_24":214,"_215":3286,"_217":3287,"_219":220},{"_212":3258,"_24":214,"_215":3259,"_217":3260,"_219":220},{"_212":3233,"_24":214,"_215":3234,"_217":3235,"_219":220},{"_212":3226,"_24":214,"_215":3227,"_217":3228,"_219":598},{"_212":3219,"_24":214,"_215":3220,"_217":3221,"_219":220},{"_212":3208,"_24":214,"_215":3209,"_217":3210,"_219":220},{"_212":3176,"_24":214,"_215":3177,"_217":3178,"_219":220},{"_212":3169,"_24":214,"_215":3170,"_217":3171,"_219":220},{"_212":3162,"_24":214,"_215":3163,"_217":3164,"_219":377},{"_212":3155,"_24":214,"_215":3156,"_217":3157,"_219":377},{"_212":3148,"_24":214,"_215":3149,"_217":3150,"_219":377},{"_212":3141,"_24":214,"_215":3142,"_217":3143,"_219":377},{"_212":3134,"_24":214,"_215":3135,"_217":3136,"_219":220},{"_212":3123,"_24":214,"_215":3124,"_217":3125,"_219":220},{"_212":3112,"_24":214,"_215":3113,"_217":3114,"_219":220},{"_212":3067,"_24":214,"_215":3068,"_217":3069,"_219":220},{"_212":3060,"_24":214,"_215":3061,"_217":3062,"_219":598},{"_212":3053,"_24":214,"_215":3054,"_217":3055,"_219":220},{"_212":3046,"_24":214,"_215":3047,"_217":3048,"_219":2773},{"_212":3025,"_24":214,"_215":3026,"_2702":2703,"_2704":2705,"_217":3027,"_219":220},{"_212":3015,"_24":214,"_215":3016,"_2702":2703,"_2704":2705,"_217":3017,"_219":220},{"_212":2993,"_24":214,"_215":2994,"_2702":2703,"_2704":2705,"_217":2995,"_219":220},{"_212":2986,"_24":214,"_215":2987,"_217":2988,"_219":220},{"_212":2979,"_24":214,"_215":2980,"_217":2981,"_219":2773},{"_212":2958,"_24":214,"_215":2959,"_2702":2703,"_2704":2705,"_217":2960,"_219":220},{"_212":2948,"_24":214,"_215":2949,"_2702":2703,"_2704":2705,"_217":2950,"_219":220},{"_212":2938,"_24":214,"_215":2939,"_2702":2703,"_2704":2705,"_217":2940,"_219":220},{"_212":2920,"_24":214,"_215":2921,"_217":2922,"_219":220},{"_212":2913,"_24":214,"_215":2914,"_217":2915,"_219":2773},{"_212":2893,"_24":214,"_215":2894,"_2702":2703,"_2704":2705,"_217":2895,"_219":220},{"_212":2872,"_24":214,"_215":2873,"_2702":2703,"_2704":2705,"_217":2874,"_219":220},{"_212":2862,"_24":214,"_215":2863,"_2702":2703,"_2704":2705,"_217":2864,"_219":220},{"_212":2855,"_24":214,"_215":2856,"_217":2857,"_219":220},{"_212":2848,"_24":214,"_215":2849,"_217":2850,"_219":2773},{"_212":2827,"_24":214,"_215":2828,"_2702":2703,"_2704":2705,"_217":2829,"_219":220},{"_212":2806,"_24":214,"_215":2807,"_2702":2703,"_2704":2705,"_217":2808,"_219":220},{"_212":2785,"_24":214,"_215":2786,"_2702":2703,"_2704":2705,"_217":2787,"_219":220},{"_212":2778,"_24":214,"_215":2779,"_217":2780,"_219":220},{"_212":2770,"_24":214,"_215":2771,"_217":2772,"_219":2773},{"_212":2748,"_24":214,"_215":2749,"_2702":2703,"_2704":2705,"_217":2750,"_219":220},{"_212":2726,"_24":214,"_215":2727,"_2702":2703,"_2704":2705,"_217":2728,"_219":220},{"_212":2700,"_24":214,"_215":2701,"_2702":2703,"_2704":2705,"_217":2706,"_219":220},{"_212":2693,"_24":214,"_215":2694,"_217":2695,"_219":220},{"_212":2686,"_24":214,"_215":2687,"_217":2688,"_219":220},"b1115aef2063",[2689],[],{"_212":2690,"_24":233,"_234":2691,"_236":2692},"cc39ba41ca8f",[],"\n","d5296f448a9f",[2696],[],{"_212":2697,"_24":233,"_234":2698,"_236":2699},"8d1da74a6c00",[],"Mubit’s system is able to prioritize procedural guidance over raw history using typing and outcome signals. This is a good benefit that depends on reliable attribution, since even a successful deployment doesn’t necessarily prove that every retrieved lesson contributed to it.","828f0b6c00db",[2710,2711,2712,2713],"level",1,"listItem","bullet",[2707],{"_212":2708,"_24":1893,"_1894":2709},"b7f12682eee8","https://docs.mubit.ai/concepts/memory-model",{"_212":2723,"_24":233,"_234":2724,"_236":2725},{"_212":2720,"_24":233,"_234":2721,"_236":2722},{"_212":2717,"_24":233,"_234":2718,"_236":2719},{"_212":2714,"_24":233,"_234":2715,"_236":2716},"cf1e8502645e",[]," includes facts, traces, lessons, rules, workflows, and exact archives, with a separate, mutable working state (rather than specifying a particular database backend).","a124afaedcbc",[2708],"memory model","c220cbdd3ded",[]," Mubit’s documented ","a1974bc579f7",[1944],"Store:","1ed494cdb84c",[2732,2733,2734,2735],[2729],{"_212":2730,"_24":1893,"_1894":2731},"03f6430bd2bb","https://docs.mubit.ai/concepts/retrieval-semantics",{"_212":2745,"_24":233,"_234":2746,"_236":2747},{"_212":2742,"_24":233,"_234":2743,"_236":2744},{"_212":2739,"_24":233,"_234":2740,"_236":2741},{"_212":2736,"_24":233,"_234":2737,"_236":2738},"82dc20558a47",[]," to combines semantic, lexical and temporal signals with working state and lesson overlays. The platform ranks information based on outcome history, and also has a direct mode that returns evidence without model-based routing or answer synthesis.","b175454eee20",[2730],"recall queries","4e86e92fe9b1",[]," The platform uses ","4390006d4807",[1944],"Read:","7f8a916d5d27",[2754,2755,2756,2757],[2751],{"_212":2752,"_24":1893,"_1894":2753},"1db8c126f415","https://docs.mubit.ai/concepts/learning-loop",{"_212":2767,"_24":233,"_234":2768,"_236":2769},{"_212":2764,"_24":233,"_234":2765,"_236":2766},{"_212":2761,"_24":233,"_234":2762,"_236":2763},{"_212":2758,"_24":233,"_234":2759,"_236":2760},"ab2045bc4caf",[]," that attaches outcomes to memory identifiers and uses reflection to derive conditional lessons for validation and promotion.","60a47075eb73",[2752],"learning loop","36a26a2ad674",[]," Mubit uses asynchronous ingestion to classify typed entries. The platform has a ","4a28c2e300ff",[1944],"Write:","23586738a3bd",[2774],[],"h3",{"_212":2775,"_24":233,"_234":2776,"_236":2777},"55a06b57abcf",[],"Mubit","66730d013fe4",[2781],[],{"_212":2782,"_24":233,"_234":2783,"_236":2784},"380b861a865e",[],"cognee’s approach lets teams customize the process of transforming source data to searchable memory. However, the approach also uses multiple derived representations, which creates extra maintenance overhead. Specifically, failed jobs can leave indexes incomplete which means not every dependent record instantly auto-updates with any corrections.","f777badb59d5",[2791,2792,2793,2794],[2788],{"_212":2789,"_24":1893,"_1894":2790},"0b959dcc7d95","https://docs.cognee.ai/core-concepts/architecture",{"_212":2804,"_24":233,"_234":2805,"_236":2725},{"_212":2801,"_24":233,"_234":2802,"_236":2803},{"_212":2798,"_24":233,"_234":2799,"_236":2800},{"_212":2795,"_24":233,"_234":2796,"_236":2797},"98a98865c12c",[]," combines relational records (for source metadata and provenance), vectors (for semantic search), and a graph (for entities and relationships).","8006ac452a9b",[2789],"architecture","b1af73cfdbec",[]," cognee’s ","3a786e68b48e",[1944],"0c8ba15851e3",[2812,2813,2814,2815],[2809],{"_212":2810,"_24":1893,"_1894":2811},"0d0d58abefcd","https://docs.cognee.ai/core-concepts/main-operations/recall",{"_212":2825,"_24":233,"_234":2826,"_236":2747},{"_212":2822,"_24":233,"_234":2823,"_236":2824},{"_212":2819,"_24":233,"_234":2820,"_236":2821},{"_212":2816,"_24":233,"_234":2817,"_236":2818},"e398196f7eb2",[]," supports different modes according to the evidence/answer required.","a7f71645af37",[2810],"recall interface","b28cafe0f3f1",[]," cognee’s retrieval uses vectors and graph structure. The ","f045c19eb548",[1944],"dcfbde86d09f",[2833,2834,2835,2836],[2830],{"_212":2831,"_24":1893,"_1894":2832},"4171b8028864","https://docs.cognee.ai/core-concepts/main-operations/improve",{"_212":2846,"_24":233,"_234":2847,"_236":2769},{"_212":2843,"_24":233,"_234":2844,"_236":2845},{"_212":2840,"_24":233,"_234":2841,"_236":2842},{"_212":2837,"_24":233,"_234":2838,"_236":2839},"5331b3438bf8",[]," that incorporate useful session information into the permanent graph.","dab6e9a6f8ee",[2831],"improvement passes","0e9444b094e5",[]," cognee uses configurable pipelines to ingest data, extract entities and relationships, and enrich existing memory. Later, the platform can also run ","9a40e96ce730",[1944],"844cbdc9c89b",[2851],[],{"_212":2852,"_24":233,"_234":2853,"_236":2854},"004318b7238a",[],"cognee","a9dca42167a7",[2858],[],{"_212":2859,"_24":233,"_234":2860,"_236":2861},"abf37fd05bec",[],"Graphiti’s approach to resolving changes during ingestion makes applicable state more explicit at read time. The trade-off here is that correct extraction becomes much more important. If the agent merges distinct workers or invalidates the wrong relationship, it may also corrupt subsequent retrieval.","c89ee678066d",[2865,2866],[],{"_212":2870,"_24":233,"_234":2871,"_236":2725},{"_212":2867,"_24":233,"_234":2868,"_236":2869},"1b9a86a13f83",[]," Graphiti uses a variety of artifacts for storage. Episodes retain source evidence, graph nodes represent entities, and edges represent relationships with temporal validity. (It should be noted that Graphiti is the open-source framework and Zep provides managed context-graph infrastructure.)","b05b70c9aa79",[1944],"3417dd3e4105",[2878,2879,2880,2881],[2875],{"_212":2876,"_24":1893,"_1894":2877},"bb5a4597ff40","https://help.getzep.com/graphiti/working-with-data/searching",{"_212":2891,"_24":233,"_234":2892,"_236":2747},{"_212":2888,"_24":233,"_234":2889,"_236":2890},{"_212":2885,"_24":233,"_234":2886,"_236":2887},{"_212":2882,"_24":233,"_234":2883,"_236":2884},"a5b1d580b210",[]," approach that combines semantic and BM25 retrieval with rank fusion. It also offers configurable recipes, which add graph traversal and node-distance ranking/cross-encoder reranking.","c45b531db105",[2876],"hybrid search","28d2a41d4e5d",[]," The platform uses a ","788181f131ab",[1944],"200e943d74d2",[2899,2900,2901,2902],[2896],{"_212":2897,"_24":1893,"_1894":2898},"44a39f9fb87c","https://github.com/getzep/graphiti",{"_212":2911,"_24":233,"_234":2912,"_236":2769},{"_212":2909,"_24":233,"_234":2910,"_236":240},{"_212":2906,"_24":233,"_234":2907,"_236":2908},{"_212":2903,"_24":233,"_234":2904,"_236":2905},"bd37a5647c0f",[]," extracts entities and relationships from incoming episodes. The platform can preserve evidence history but allows new evidence to invalidate previous relationships.","646ac358d278",[2897],"Graphiti","2bf68a99ad63",[],"23ab9ad7f341",[1944],"ec1a6ae1873c",[2916],[],{"_212":2917,"_24":233,"_234":2918,"_236":2919},"4d803c10905d",[],"Graphiti and Zep","27fdcaf80cf2",[2926,2927,2928],[2923],{"_212":2924,"_24":1893,"_1894":2925},"39e749b78ea5","https://github.com/mem0ai/mem0",{"_212":2935,"_24":233,"_234":2936,"_236":2937},{"_212":2932,"_24":233,"_234":2933,"_236":2934},{"_212":2929,"_24":233,"_234":2930,"_236":2931},"de1e3d5a2c9a",[]," that differs from the managed architecture described above.)","e5f17f165be9",[2924],"open-source implementation","95b36f2da3c8",[],"Mem0’s approach uses a standardized pipeline to interpret information during an agentic run. The platform takes the approach of preserving evidence by appending changed facts, but it leaves version resolution up to retrieval and reader. (Mem0 also publishes an ","98d3de1d13c4",[2941,2942],[],{"_212":2946,"_24":233,"_234":2947,"_236":2725},{"_212":2943,"_24":233,"_234":2944,"_236":2945},"3cc6a355c602",[]," Memory text, embeddings, and metadata live in a vector database. Mem0 uses an entity store to link related memories, and retains additional history and a rolling message window using SQL.","0dfb36a8c102",[1944],"39c6e15e2909",[2951,2952],[],{"_212":2956,"_24":233,"_234":2957,"_236":2747},{"_212":2953,"_24":233,"_234":2954,"_236":2955},"1711f15ec7f4",[]," Mem0 ranks information relevance based on semantic, BM25, entity, and temporal signals. Temporal relevance ranks information by freshness rather than immediately excluding stale candidates.","99f313c464e3",[1944],"6625977f2d5a",[2964,2965,2966,2967],[2961],{"_212":2962,"_24":1893,"_1894":2963},"d6a326566adf","https://docs.mem0.ai/core-concepts/memory-evaluation",{"_212":2977,"_24":233,"_234":2978,"_236":2769},{"_212":2974,"_24":233,"_234":2975,"_236":2976},{"_212":2971,"_24":233,"_234":2972,"_236":2973},{"_212":2968,"_24":233,"_234":2969,"_236":2970},"6d3effc41325",[]," processes conversations asynchronously, extracts ADD-only facts, deduplicates them, creates embeddings, and adds entity links and temporal metadata.","60b128b85711",[2962],"extraction pipeline","724c2c891981",[]," Mem0’s ","5a484c730a59",[1944],"813631713196",[2982],[],{"_212":2983,"_24":233,"_234":2984,"_236":2985},"d9f878ff27a0",[],"Mem0's Managed Platform","031f13e60a30",[2989],[],{"_212":2990,"_24":233,"_234":2991,"_236":2992},"7b3d106bb226",[],"Letta’s approach gives capable models flexibility over both representation and retrieval. Because Letta allows for persistent memory editing and on-demand retrieval, the correctness of a job depends on editing and search decisions. Version control provides rollback capability, but does not validate the meaning of a memory update.","0aaa75402c19",[2999,3000,3001,3002],[2996],{"_212":2997,"_24":1893,"_1894":2998},"bab345521546","https://www.letta.com/blog/memory-blocks/",{"_212":3012,"_24":233,"_234":3013,"_236":3014},{"_212":3009,"_24":233,"_234":3010,"_236":3011},{"_212":3006,"_24":233,"_234":3007,"_236":3008},{"_212":3003,"_24":233,"_234":3004,"_236":3005},"28d8492c3fe9",[],", which persist labeled, size-limited text. Context repositories represent memory as versioned files, with designated files loaded into the prompt.","2e258a11f020",[2997,1926],"memory blocks","bd33ca7d5b2b",[],"Letta abstracts context window storage as ","3a82d6616947",[1944],"Store: ","f63a7efb1736",[3018,3019],[],{"_212":3023,"_24":233,"_234":3024,"_236":2747},{"_212":3020,"_24":233,"_234":3021,"_236":3022},"c6ee02242da8",[]," Letta uses a combination of automatically including records and using agents to access additional memory through tools. In context repositories, the file hierarchy helps it discover material to load on demand.","225fb3f5614e",[1944],"d1ca287a4e65",[3031,3032,3033,3034],[3028],{"_212":3029,"_24":1893,"_1894":3030},"d611832c5bfa","https://www.letta.com/blog/context-repositories/",{"_212":3044,"_24":233,"_234":3045,"_236":2769},{"_212":3041,"_24":233,"_234":3042,"_236":3043},{"_212":3038,"_24":233,"_234":3039,"_236":3040},{"_212":3035,"_24":233,"_234":3036,"_236":3037},"b30686ec4f7d",[]," let memory agents work in isolated Git worktrees and merge their changes.","2162f1f7e868",[3029],"context repositories","7f1857d4cccd",[]," Agents edit persistent memory using tools, with background agents handling consolidation. Letta Code's ","e107ebffb775",[1944],"0893565179df",[3049],[],{"_212":3050,"_24":233,"_234":3051,"_236":3052},"606ad4cb319c",[],"Letta and Letta Code","4f649d90264a",[3056],[],{"_212":3057,"_24":233,"_234":3058,"_236":3059},"245d14e460fa",[],"Below are some real-world examples of how different platforms approach agentic memory, at least, according to publicly-available source code or architecture explainers:","4747fa21e7a7",[3063],[],{"_212":3064,"_24":233,"_234":3065,"_236":3066},"a0bf90cc4808",[],"How Existing Systems Combine These Choices","26e03d89cef4",[3076,3077,3078,3079,3080,3081,3082,3083,3084],[3070,3071],{"_212":3074,"_24":1893,"_1894":3075},{"_212":3072,"_24":1893,"_1894":3073},"c09991de458c","https://arxiv.org/abs/2512.13564","580336df6328","https://www.ibm.com/docs/en/db2/12.1.x?topic=indexes-vector",{"_212":3109,"_24":233,"_234":3110,"_236":3111},{"_212":3106,"_24":233,"_234":3107,"_236":3108},{"_212":3103,"_24":233,"_234":3104,"_236":3105},{"_212":3100,"_24":233,"_234":3101,"_236":3102},{"_212":3097,"_24":233,"_234":3098,"_236":3099},{"_212":3094,"_24":233,"_234":3095,"_236":3096},{"_212":3091,"_24":233,"_234":3092,"_236":3093},{"_212":3088,"_24":233,"_234":3089,"_236":3090},{"_212":3085,"_24":233,"_234":3086,"_236":3087},"2fbdca59fcea",[],".)","c88f2f3993b8",[3072],"Memory in the Age of AI Agents","241c4055abee",[]," determine whether facts are current, whether records refer to the same worker, or whether a proposed lesson is justified. For example, selection, organization, and retrieval seem like a huge part of memory design for Claude Code, even when the durable records are ordinary files. (For more on latent and parametric memory, see the report ","592bc7c5d608",[1926],"not","f69b2c969e9a",[],"different representations to find similar content. However, vector indexes do ","51dc1c62d004",[1926],"alongside ","4f49328c1072",[]," (specialized data structures that organize embeddings to speed up searches) can sit ","bfff0f664909",[3074,1926],"Vector indexes","65b134718ba7",[1926],"Note: ","459ca3e10e7d",[3115,3116],[],{"_212":3120,"_24":233,"_234":3121,"_236":3122},{"_212":3117,"_24":233,"_234":3118,"_236":3119},"9580a2df3917",[]," Using graphs could connect customers to worker fleets, fleets to software versions, and versions to schema requirements. A key feature of graphs is temporal relationships, which could indicate when each dependency applied during the run. With graphs, multi-step questions (like which customers still depend on a particular column) become easier to express. However, as underlying deployments change, graphs require extraction, entity resolution, and updates to remain accurate.","6597cc79ccc4",[1944],"Entity-relationship graphs.","91041a2f7d82",[3126,3127],[],{"_212":3131,"_24":233,"_234":3132,"_236":3133},{"_212":3128,"_24":233,"_234":3129,"_236":3130},"048b0f97261a",[]," If you need to answer questions across many customers and deployments, another approach is using rows or documents for important fields like customer_id, worker_version, observed_at, and source_event_id. Using fields lets agents filter and make explicit version checks. However, you would need a schema that tracks version changes among workers, which is a distinction that can be lost in a field overwrite unless you also retain earlier versions.","c4107be75da4",[1944],"Structured records.","f32fca39f7ed",[3137],[],{"_212":3138,"_24":233,"_234":3139,"_236":3140},"cfea8479c38f",[],"The index helps discover the notes, which explain the constraint and point to its source. Using ordinary file tools makes such summaries easy to inspect and revise. Also, files can contain structured metadata, even if our system chooses to use Markdown. The challenge here is consistency, since notes can be readable while contradicting each other. Adding versioning makes changes reversible, but does not validate them. (See Letta Code, below, for an example.)","06c83b269941",[3144],[],{"_212":3145,"_24":233,"_234":3146,"_236":3147},"aa317516e315",[],"└── reference_customer_rollout_tracker.md","e6d70c556e93",[3151],[],{"_212":3152,"_24":233,"_234":3153,"_236":3154},"1b55f6b25b56",[],"├── feedback_migration_compatibility.md","9d066e09a773",[3158],[],{"_212":3159,"_24":233,"_234":3160,"_236":3161},"baf15381d1a5",[],"├── MEMORY.md","92b2b15e6ef5",[3165],[],{"_212":3166,"_24":233,"_234":3167,"_236":3168},"593cd17fb5cc",[],"memory/","989c53d303db",[3172],[],{"_212":3173,"_24":233,"_234":3174,"_236":3175},"9091690fe872",[],"Here is an illustrative layout for our example:","586a0cf2fe77",[3185,3186,3187,3188,3189,3190],[3179,3180],{"_212":3183,"_24":1893,"_1894":3184},{"_212":3181,"_24":1893,"_1894":3182},"bf2bc5254bf0","https://code.claude.com/docs/en/memory#storage-location","961db163e8ee","https://code.claude.com/docs/en/skills",{"_212":3205,"_24":233,"_234":3206,"_236":3207},{"_212":3202,"_24":233,"_234":3203,"_236":3204},{"_212":3199,"_24":233,"_234":3200,"_236":3201},{"_212":3196,"_24":233,"_234":3197,"_236":3198},{"_212":3193,"_24":233,"_234":3194,"_236":3195},{"_212":3191,"_24":233,"_234":3192,"_236":2543},"84bc558b7fb4",[],"9b5977a7be50",[3181],"default memory is local to the machine","8411ca9c5db5",[],", but the ","3dbe26891ce6",[3183,1926],"frontmatter","0f9e248beec2",[]," Claude Code stores auto memory as Markdown under ~/.claude/projects//memory/, with a MEMORY.md index and individual topic files. Different memory types appear in ","63ac9cbc656e",[1944],"Text files and summaries.","221d0943bbbe",[3211,3212],[],{"_212":3216,"_24":233,"_234":3217,"_236":3218},{"_212":3213,"_24":233,"_234":3214,"_236":3215},"7c4441ab063c",[]," Claude Code's session transcripts show that the model retains interactions. (Files or object storage can also hold command output, patches, etc..) Such artifacts help agents investigate and reprocess, but even a full history does not guarantee you can immediately identify which constraint applies to the current job. For that, you would need your agents to run a subsequent read or derived view.","8961e61692ba",[1944],"Raw logs and artifacts.","4ce4eb451d82",[3222],[],{"_212":3223,"_24":233,"_234":3224,"_236":3225},"07928571dd78",[],"There are multiple ways to represent an agent’s experience, including an execution record, a written lesson, a structured fact, or a relationship. Your choice determines what aspects of an agentic job you can preserve and query directly. (Your backend determines how records persist and how to coordinate concurrent access.) We’ll cover four approaches below:","1adfa149433a",[3229],[],{"_212":3230,"_24":233,"_234":3231,"_236":3232},"788c87082178",[],"Storage: Representation, Backend, and Indexes","7d7e734abb9d",[3242,3243,3244,3245],[3236,3237],{"_212":3240,"_24":1893,"_1894":3241},{"_212":3238,"_24":1893,"_1894":3239},"067a18a9a1e9","https://code.claude.com/docs/en/how-claude-code-works#when-context-fills-up","1af973a49559","https://learn.microsoft.com/en-us/agent-framework/concepts/agents/conversations/compaction",{"_212":3255,"_24":233,"_234":3256,"_236":3257},{"_212":3252,"_24":233,"_234":3253,"_236":3254},{"_212":3249,"_24":233,"_234":3250,"_236":3251},{"_212":3246,"_24":233,"_234":3247,"_236":3248},"856e85a5c3db",[],", supporting continuity within a session while treating the selection of durable lessons for future sessions as a separate decision.","66110f9314ae",[3238],"summarize a growing conversation within its context window","49dd1e323c26",[]," (condensing older conversation history, reasoning, and tool outputs to conserve context window) can also help here. Claude Code’s approach is to ","79c644312837",[3240,1926],"Compaction","1d284f0bdd3e",[3261,3262,3263,3264,3265,3266],[],{"_212":3282,"_24":233,"_234":3283,"_236":3284},{"_212":3279,"_24":233,"_234":3280,"_236":3281},{"_212":3276,"_24":233,"_234":3277,"_236":3278},{"_212":3273,"_24":233,"_234":3274,"_236":3275},{"_212":3270,"_24":233,"_234":3271,"_236":3272},{"_212":3267,"_24":233,"_234":3268,"_236":3269},"b5f67f56504d",[],". It’s possible for both automatic or agentic-requested read requests to use exact lookups, keyword search, vector similarity, or graph traversal. (Harnesses can also combine policies like including a small index while auto-retrieving known task constraints and letting agents investigate further.) The goal is to select enough evidence for decision-making without having to store the entire history into every call.","faf7027e81fd",[1926],"how retrieval works","17b0e32a878e",[]," initiates retrieval is separate from ","9c22117de48b",[1926],"Who","d277a5a24136",[],": ","8d32937852d7",[1926],"About retrieval and memory overhead","baec8a07be2b",[3291,3292,3293,3294],[3288],{"_212":3289,"_24":1893,"_1894":3290},"0f84a55baee9","https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents",{"_212":3304,"_24":233,"_234":3305,"_236":3306},{"_212":3301,"_24":233,"_234":3302,"_236":3303},{"_212":3298,"_24":233,"_234":3299,"_236":3300},{"_212":3295,"_24":233,"_234":3296,"_236":3297},"413c9af36bc8",[],", so the model can discover a clue, refine its question, and perform additional retrieval. Agents using this policy might start by reading a rollout note, follow its reference to the tracker, and check current worker versions before proposing a migration (or, if the task was about why the original migration failed, it would search retained session records for more details). This approach adds flexibility but also adds costs for model calls and tool use. Having a compact index can help an investigation start better, but isn’t a guarantee it will end well, as agents may fail to search, choose the wrong note, or stop short of checking the source.","bfef448b405c",[3289],"Claude Code combines upfront context with on-demand retrieval","de9d369d54d7",[]," Another approach is to let agents use tools to investigate. For example, ","5db73706b4b4",[1944],"Let the model retrieve memory as needed.","3a95128b53ea",[3310,3311,3312,3313,3314,3315],[],{"_212":3331,"_24":233,"_234":3332,"_236":3333},{"_212":3328,"_24":233,"_234":3329,"_236":3330},{"_212":3325,"_24":233,"_234":3326,"_236":3327},{"_212":3322,"_24":233,"_234":3323,"_236":3324},{"_212":3319,"_24":233,"_234":3320,"_236":3321},{"_212":3316,"_24":233,"_234":3317,"_236":3318},"4f39fb880af4",[],". However, this policy carries the risk that the harness might introduce new issues (like constructing a poor query, returning stale records, or filling context with superficially similar stuff).","1c9c00d144c4",[1926],"think to search","d71331d31241",[]," selected model calls, without waiting for the model to request it. This policy requires a migration-aware harness to proactively identify any affected services/schemas, retrieve any relevant deployment constraints, and attach them to the next call. (This approach would support exact filters with structured records and allow free-form notes to be text-searched.) The benefit here is that retrieval is not constrained by whether agents ","58a66df2bcde",[1926],"before","8406b79505b0",[]," Another approach runs retrieval ","7061a6e00d28",[1944],"Have the harness retrieve and assemble context.","da3b59765f18",[3337],[],{"_212":3338,"_24":233,"_234":3339,"_236":3340},"8a34dbf8511d",[],"Automatic inclusion makes discovery more predictable, but consumes context even on unrelated tasks. It also requires a choice about what deserves that space. CLAUDE.md supplies persistent instructions; auto memory contains Claude's own notes. Both guide behavior rather than enforce it. A deployment constraint that must be guaranteed needs an execution check as well as a reminder. Claude Code's documentation explains this distinction.","5ca9d96310af",[3350,3351,3352,3353,3354,3355,3356,3357,3358,3359],[3344,3345],{"_212":3348,"_24":1893,"_1894":3349},{"_212":3346,"_24":1893,"_1894":3347},"2fe7d1e0fb7c","https://code.claude.com/docs/en/memory#claudemd-vs-auto-memory","6b5816824a71","https://code.claude.com/docs/en/memory#how-it-works",{"_212":3387,"_24":233,"_234":3388,"_236":3389},{"_212":3384,"_24":233,"_234":3385,"_236":3386},{"_212":3381,"_24":233,"_234":3382,"_236":3383},{"_212":3378,"_24":233,"_234":3379,"_236":3380},{"_212":3375,"_24":233,"_234":3376,"_236":3377},{"_212":3372,"_24":233,"_234":3373,"_236":3374},{"_212":3369,"_24":233,"_234":3370,"_236":3371},{"_212":3366,"_24":233,"_234":3367,"_236":3368},{"_212":3363,"_24":233,"_234":3364,"_236":3365},{"_212":3360,"_24":233,"_234":3361,"_236":3362},"f04028db5006",[]," between persistent instructions vs. auto memory.)","72cea0fbced3",[3346],"Claude Code distinguishes","0ca414d6225d",[],"because a concise and descriptive entry makes the right notes easier to discover (and bad descriptions make it harder). This approach makes discovery more predictable but consumes context on unrelated tasks, and requires a choice about what is worth storing. (For reference, ","0301e86aad6f",[1926],"write ","a7474d19202d",[]," with ","d9ba8f226466",[1926],"read","53bf244436b4",[],"). Here, record storage does not count against total stored memory, as any reading of detailed topic files is done on demand. This approach connects ","858d0ef5328a",[3348],"loads MEMORY.md at conversation start","351d6bf3467b",[]," One approach is to keep a small set of instructions or pointers within context (Claude Code ","81afda08f68e",[1944],"Include designated records automatically.","0d2f5b0ff899",[3393],[],{"_212":3394,"_24":233,"_234":3395,"_236":3396},"b87351817f0c",[],"After noting changes from their write policy, agents need to understand whether written changes affect them and any upgrades they have received. These three read policies are most common:","330c4b5540cf",[3400],[],{"_212":3401,"_24":233,"_234":3402,"_236":3403},"978e8914fa68",[],"Read Policy: What Reaches the Model?","28f6819109be",[3407,3408,3409,3410],[],{"_212":3419,"_24":233,"_234":3420,"_236":3421},{"_212":3416,"_24":233,"_234":3417,"_236":3418},{"_212":3414,"_24":233,"_234":3415,"_236":3096},{"_212":3411,"_24":233,"_234":3412,"_236":3413},"17bde65913e4",[]," mutually exclusive. For example, you could retain an exchange, add a brief reference, and reconcile it with other experiences later. Of course, you would need to account for delays if subsequent processing happens in the background. Conversely, you would have to account for newly-discovered constraints if you ran immediate retries.","85f7c1ab0cff",[1926],"2635523039da",[]," These write policies are ","fcd21325e323",[1926],"Note:","5861fa445fa5",[3433,3434,3435,3436,3437,3438,3439,3440,3441,3442,3443,3444],[3425,3426,3427],{"_212":3431,"_24":1893,"_1894":3432},{"_212":3429,"_24":1893,"_1894":3430},{"_212":3428,"_24":1893,"_1894":3030},"cf05d9d85127","94e81cff0ae0","https://www.ibm.com/docs/en/informix-servers/12.10.0?topic=modeling-fact-table","b1a64c40a930","https://en.wikipedia.org/wiki/Data_extraction",{"_212":3478,"_24":233,"_234":3479,"_236":3480},{"_212":3475,"_24":233,"_234":3476,"_236":3477},{"_212":3472,"_24":233,"_234":3473,"_236":3474},{"_212":3469,"_24":233,"_234":3470,"_236":3471},{"_212":3466,"_24":233,"_234":3467,"_236":3468},{"_212":3463,"_24":233,"_234":3464,"_236":3465},{"_212":3460,"_24":233,"_234":3461,"_236":3462},{"_212":3457,"_24":233,"_234":3458,"_236":3459},{"_212":3454,"_24":233,"_234":3455,"_236":3456},{"_212":3451,"_24":233,"_234":3452,"_236":3453},{"_212":3448,"_24":233,"_234":3449,"_236":3450},{"_212":3445,"_24":233,"_234":3446,"_236":3447},"91234870a812",[]," for documented examples of using background consolidation to update agent memory.","938fcb5e36a8",[3428],"Letta Code's context repositories","b37a5929e4ab",[]," (deciding whether new facts should update an existing memory). See ","5e3b514ea686",[1926],"consolidation","cdd413341cbd",[],") from ","f07d4da484cd",[3429],"candidate facts","00c26c30913f",[]," (used to identify ","ce486f61cec6",[1926],"extraction","8f6092b664db",[],". Converting the resulting data from the job into appropriate data fields provides a consistent schema and lets us rerun extraction when the policy changes. However, the extra processing adds cost and potential risk of misinterpreting evidence. It also separates ","efeadcbf2b1c",[3431,1926],"extractor","c729269fd4b9",[]," Another approach is processing recorded sessions independently of the agent with an ","097ba3fb5001",[1944],"Use a separate extraction pipeline.","b0388f909a33",[3487,3488,3489,3490,3491,3492],[3484],{"_212":3485,"_24":1893,"_1894":3486},"86369451d645","https://code.claude.com/docs/en/memory#auto-memory",{"_212":3508,"_24":233,"_234":3509,"_236":3510},{"_212":3505,"_24":233,"_234":3506,"_236":3507},{"_212":3502,"_24":233,"_234":3503,"_236":3504},{"_212":3499,"_24":233,"_234":3500,"_236":3501},{"_212":3496,"_24":233,"_234":3497,"_236":3498},{"_212":3493,"_24":233,"_234":3494,"_236":3495},"ed7d89ec8a3e",[],"saving (such as which data column changed), while the operating constraint, and where to verify it, becomes durable. This approach requires less reconstruction on the next run. However, there’s a risk of misinterpreting a one-off issue, like a customer delay, as a hard-and-fast rule to implement (which can be mitigated with notes/references to supporting evidence).","58fd06e03e07",[1926],"worth ","a377237b09e8",[]," selectively saves user info, feedback, context, and references, skipping facts recoverable from code or Git history. The difference is what is ","ea35d09d64e9",[3485],"auto memory","d0750afe804b",[]," Claude Code's ","a3d9b929d5ec",[1944],"Let the active agent write a memory.","d809f9f408ed",[3517,3518,3519,3520],[3514],{"_212":3515,"_24":1893,"_1894":3516},"f5a4020c4fb9","https://code.claude.com/docs/en/how-claude-code-works#work-with-sessions",{"_212":3530,"_24":233,"_234":3531,"_236":3532},{"_212":3527,"_24":233,"_234":3528,"_236":3529},{"_212":3524,"_24":233,"_234":3525,"_236":3526},{"_212":3521,"_24":233,"_234":3522,"_236":3523},"c0a92cd924e3",[],". Recording means preserving a failed query + investigation + any dev explanation, but does not require immediate decisions on which details matter. Auto-capture moves the cost of the write policy to read, as someone will need to search for and interpret it later (which would mean that this approach also requires retention).","de3b99a10fc3",[3515],"records messages, tool calls, and results","3c08d4da140a",[]," Claude Code ","d53459c4b154",[1944],"Automatically capture execution.","ea5d8e984d9d",[3536],[],{"_212":3537,"_24":233,"_234":3538,"_236":3539},"6cedcb2c785b",[],"A write policy determines what gets retained, who interprets it, and when the result becomes available. These three are most common:","ccf4da4770a3",[3543],[],{"_212":3544,"_24":233,"_234":3545,"_236":3546},"424c91921778",[],"Write Policy: What Becomes Memory?","d676ce5bde85",[3550],[],{"_212":3551,"_24":233,"_234":3552,"_236":3553},"94f5b2979df1",[],"To follow these choices through one task, imagine using Claude Code to investigate a database migration that broke an older worker. During the investigation, the developer explains that customers upgrade their workers independently of the backend, and that a rollout tracker records which versions are still running. What should survive it, and how should that information affect the next migration?","26defa59f442",[3560,3561,3562],[3557],{"_212":3558,"_24":1893,"_1894":3559},"e04bb9c15fd2","https://x.com/himanshustwts/status/2038924027411222533",{"_212":3569,"_24":233,"_234":3570,"_236":3571},{"_212":3566,"_24":233,"_234":3567,"_236":3568},{"_212":3563,"_24":233,"_234":3564,"_236":3565},"921f830b5e17",[1926],". This is a source-code analysis, not an official architecture diagram. Internal mechanisms such as autoDream and extractMemories reflect the version analyzed. The walkthrough below uses public documentation for shipped behavior.","9767f044ab49",[3558,1926],"himanshu on X, March 31, 2026","083a41cf889e",[1926],"Source: ","ad95b655bc4d",{"_1874":3574,"_24":1876},"image-f8350e46e885d0b2fa3ac0a247b0dbef59722d0b-1200x720-jpg","6196452f03db",[3580,3581,3582],[3578],{"_212":3579,"_24":1893,"_1894":3559},"b6eb1cd451c2",{"_212":3589,"_24":233,"_234":3590,"_236":3591},{"_212":3586,"_24":233,"_234":3587,"_236":3588},{"_212":3583,"_24":233,"_234":3584,"_236":3585},"4b7379850874",[]," while reading about memory systems. The diagram separates the paths that create and revise memory from the paths that bring it back into context:","dc5e6e608e02",[3579],"himanshu's walkthrough of Claude Code's memory architecture","3a123eb6cc1c",[],"I came across ","7abfa2abb045",[3595],[],{"_212":3596,"_24":233,"_234":3597,"_236":3598},"09e9d2285318",[],"The implementation of memory comes down to three connected choices: the write policy, the read policy, and the storage representation. Claude Code gives us a concrete example of how a harness combines them.","a70d3dc5602d",[3602],[],{"_212":3603,"_24":233,"_234":3604,"_236":3605},"4a35c5e079f4",[],"F determines which observations become persistent state and how they revise existing records. R selects the state that reaches the next model call. Both can combine stochastic compute and deterministic code. Outcomes may arrive much later than actions. The write path must associate delayed evidence with the original action and revise any lessons derived from it.","631992c5db66",[3609],[],{"_212":3610,"_24":233,"_234":3611,"_236":3612},"904ce784cc6b",[],")","33e2a8d52b3b",[3616],[],{"_212":3617,"_24":233,"_234":3618,"_236":3619},"7680b5ad5684",[],"outcome_t","f4d41d0ffe96",[3623],[],{"_212":3624,"_24":233,"_234":3625,"_236":3626},"80d216e10ccb",[],"action_t,","7f8cc217016d",[3630],[],{"_212":3631,"_24":233,"_234":3632,"_236":3633},"9d5241992604",[],"observation_t,","79b91869e2e3",[3637],[],{"_212":3638,"_24":233,"_234":3639,"_236":3640},"76a995443776",[],"M_t,","f082f460dd34",[3644],[],{"_212":3645,"_24":233,"_234":3646,"_236":3647},"66125ac80fbe",[],"M_t+1 = F(","d3f764a17d9c",[3651],[],{"_212":3652,"_24":233,"_234":3653,"_236":3654},"72a3b7698b71",[],"","cbc2ed7ad74a",[3658],[],{"_212":3659,"_24":233,"_234":3660,"_236":3612},"2816fd2f0683",[],"e6879dfd579c",[3664],[],{"_212":3665,"_24":233,"_234":3666,"_236":3667},"2611c7cb1e9f",[],"observation_t","f22ddf871918",[3671],[],{"_212":3672,"_24":233,"_234":3673,"_236":3674},"affdd49b21ac",[],"context_t,","0b328b527a81",[3678],[],{"_212":3679,"_24":233,"_234":3680,"_236":3681},"2a217f0d1f95",[],"working_state_t,","fbf8e2258ec2",[3685],[],{"_212":3686,"_24":233,"_234":3687,"_236":3688},"ac90fae6dc2e",[],"instructions,","d90d632afed0",[3692],[],{"_212":3693,"_24":233,"_234":3694,"_236":3695},"f9eaeeeb18fd",[],"action_t = model(","bccb7605f45d",[3699],[],{"_212":3700,"_24":233,"_234":3701,"_236":3654},"4b7a59ffa024",[],"62cc0bf02726",[3705],[],{"_212":3706,"_24":233,"_234":3707,"_236":3708},"c20486f8ce97",[],"context_t = R(query_t, M_t, B)","8a7574d4e676",[3712],[],{"_212":3713,"_24":233,"_234":3714,"_236":3715},"6a62e755d8ee",[],"A memory-augmented agent adds persistent state M_t, a write policy F, a retrieval policy R, and a context budget B:","73cf6663ed99",[3719],[],{"_212":3720,"_24":233,"_234":3721,"_236":3722},"28413b799a65",[],"action_t = model(instructions, recent_messages, observation_t)","02f0e6bf9487",[3726],[],{"_212":3727,"_24":233,"_234":3728,"_236":3729},"057f4a37a0a5",[],"At time t, an agent receives an observation, assembles context and chooses an action. We can therefore describe a simplified agent without memory as:","607ef2412da8",[3733],[],{"_212":3734,"_24":233,"_234":3735,"_236":3736},"52f17d0d3b1a",[],"Memory as a Stateful Control Loop","cb4932e369b2",[3740,3741,3742],[],{"_212":3749,"_24":233,"_234":3750,"_236":3751},{"_212":3746,"_24":233,"_234":3747,"_236":3748},{"_212":3743,"_24":233,"_234":3744,"_236":3745},"e314c861bd9f",[]," the model, in the harness layer that controls the agent loop.","d84a78243562",[1926],"outside","778e700cc7c0",[],"It is tempting to describe this as a storage problem. Save the agent's interactions, embed them, and retrieve the closest results before the next model call. That produces persistence, but not necessarily memory. A useful memory system must decide what to write, how to represent it, when to retrieve it, how to resolve conflicting information, and whether/how an outcome should modify future behavior. Current systems implement this functionality ","121f4a444355",[3755],[],{"_212":3756,"_24":233,"_234":3757,"_236":3758},"18e76a78164b",[],"It cannot coordinate with other agents if each action begins with an isolated view of the organization","72565cf8678f",[3762],[],{"_212":3763,"_24":233,"_234":3764,"_236":3765},"744b85306b2c",[],"Learning can't happen without memory, and an agent cannot learn if the outcome of one attempt disappears before the next","a0eeaffd296d",[3769],[],{"_212":3770,"_24":233,"_234":3771,"_236":3772},"0f74cd6c9885",[],"An agent cannot pursue a goal over months if it forgets why the goal/sub-goal matters","7139f92c87c1",[3776],[],{"_212":3777,"_24":233,"_234":3778,"_236":3779},"6a42f6ec679f",[],"To understand agent memory, we shouldn't think of it as a place to store the past. We need to think of it as a harness-level control system that decides which previous experiences should affect future behavior. The need for memory in the long-horizon context should be clear, but to provide a few examples:","51c7e00b5f52",[3786,3787,3788],[3783],{"_212":3784,"_24":1893,"_1894":3785},"f72dcfadb07b","https://www.heavybit.com/library/article/long-horizon-agents-from-order-takers-to-outcome-owners",{"_212":3795,"_24":233,"_234":3796,"_236":3797},{"_212":3792,"_24":233,"_234":3793,"_236":3794},{"_212":3789,"_24":233,"_234":3790,"_236":3791},"ceb09fa7a740",[],", I laid out the five systems I believe are missing before agents can graduate from completing tasks to owning outcomes: Memory, learning, goal orchestration, a durable execution layer, and a toolchain ecosystem. This post dives into agent memory.","d6a782b5390a",[3784],"Part 1","dade4a80a130",[],"In ","829b03b465e4",[3801],[],{"_212":3802,"_24":233,"_234":3803,"_236":3654},"224565565806",[],"16e0361b4ce2",[3807],[],{"_212":3808,"_24":233,"_234":3809,"_236":3810},"991570e6c69f",[],"Why Memory Is A Key Piece of the Long-Horizon Puzzle",[3817,3818,3819,3820,3821,3822,3823,3824,3825,3826,3827,3828,3829,3830,3831,3832,3833,3834,3835,3836,3837,3838,3839,3840,3841,3842,3843,3844,3845,3846,3847,3848,3849,3850,3851,3852,3853,3854,3855,3856,3857,3858,3859,3860,3861,3862,3863,3864,3865,3866,3867],16,"616147c4-aa4e-4ab9-83ae-f20398e6871d","2026-10-20T16:48:00.000Z","how-to-orchestrate-multiple-agents-from-the-terminal","How to Orchestrate Multiple Agents from the Terminal",{"_212":4422,"_24":214,"_215":4423,"_217":4424,"_219":598},{"_212":4404,"_24":214,"_215":4405,"_217":4406,"_219":220},{"_212":4386,"_24":214,"_215":4387,"_217":4388,"_219":220},{"_212":4383,"_24":1949,"_74":4384},{"_212":4369,"_24":214,"_215":4370,"_217":4371,"_219":220},{"_212":4362,"_24":214,"_215":4363,"_217":4364,"_219":598},{"_212":4344,"_24":214,"_215":4345,"_217":4346,"_219":220},{"_212":4321,"_24":214,"_215":4322,"_217":4323,"_219":220},{"_212":4306,"_24":214,"_215":4307,"_217":4308,"_219":220},{"_212":4299,"_24":214,"_215":4300,"_217":4301,"_219":220},{"_212":4292,"_24":214,"_215":4293,"_217":4294,"_219":598},{"_212":4285,"_24":214,"_215":4286,"_217":4287,"_219":220},{"_212":4270,"_24":214,"_215":4271,"_217":4272,"_219":220},{"_212":4263,"_24":214,"_215":4264,"_217":4265,"_219":220},{"_212":4256,"_24":214,"_215":4257,"_217":4258,"_219":220},{"_212":4249,"_24":214,"_215":4250,"_217":4251,"_219":220},{"_212":4242,"_24":214,"_215":4243,"_217":4244,"_219":220},{"_212":4219,"_24":214,"_215":4220,"_217":4221,"_219":220},{"_212":4212,"_24":214,"_215":4213,"_217":4214,"_219":598},{"_212":4205,"_24":214,"_215":4206,"_217":4207,"_219":220},{"_212":4198,"_24":214,"_215":4199,"_217":4200,"_219":220},{"_212":4180,"_24":214,"_215":4181,"_217":4182,"_219":220},{"_212":4173,"_24":214,"_215":4174,"_217":4175,"_219":377},{"_212":4166,"_24":214,"_215":4167,"_217":4168,"_219":598},{"_212":4140,"_24":214,"_215":4141,"_217":4142,"_219":220},{"_212":4125,"_24":214,"_215":4126,"_217":4127,"_219":220},{"_212":4111,"_24":214,"_215":4112,"_217":4113,"_219":220},{"_212":4096,"_24":214,"_215":4097,"_217":4098,"_219":220},{"_212":4089,"_24":214,"_215":4090,"_217":4091,"_219":220},{"_212":4082,"_24":214,"_215":4083,"_217":4084,"_219":220},{"_212":4075,"_24":214,"_215":4076,"_217":4077,"_219":598},{"_212":4060,"_24":214,"_215":4061,"_217":4062,"_219":220},{"_212":4053,"_24":214,"_215":4054,"_217":4055,"_219":220},{"_212":4022,"_24":214,"_215":4023,"_217":4024,"_219":220},{"_212":4015,"_24":214,"_215":4016,"_217":4017,"_219":220},{"_212":4008,"_24":214,"_215":4009,"_217":4010,"_219":598},{"_212":4001,"_24":214,"_215":4002,"_217":4003,"_219":220},{"_212":3994,"_24":214,"_215":3995,"_217":3996,"_219":220},{"_212":3987,"_24":214,"_215":3988,"_217":3989,"_219":220},{"_212":3980,"_24":214,"_215":3981,"_217":3982,"_219":220},{"_212":3973,"_24":214,"_215":3974,"_217":3975,"_219":598},{"_212":3955,"_24":214,"_215":3956,"_217":3957,"_219":220},{"_212":3924,"_24":214,"_215":3925,"_217":3926,"_219":220},{"_212":3917,"_24":214,"_215":3918,"_217":3919,"_219":220},{"_212":3910,"_24":214,"_215":3911,"_217":3912,"_219":220},{"_212":3903,"_24":214,"_215":3904,"_217":3905,"_219":220},{"_212":3896,"_24":214,"_215":3897,"_217":3898,"_219":377},{"_212":3889,"_24":214,"_215":3890,"_217":3891,"_219":598},{"_212":3882,"_24":214,"_215":3883,"_217":3884,"_219":220},{"_212":3875,"_24":214,"_215":3876,"_217":3877,"_219":220},{"_212":3868,"_24":214,"_215":3869,"_217":3870,"_219":220},"3e791b59e928",[3871],[],{"_212":3872,"_24":233,"_234":3873,"_236":3874},"19f013dfac9b",[],"“So I don't think tools like TUIOS go away. I think the job changes. Today it helps you keep an eye on a handful of agents. Next it has to tell you when to look, show you the evidence when you do, and let you jump in at any moment. And honestly, a lot of TUIOS today gets built by agents running inside TUIOS, so I'm motivated to get that part right!”","fb18f8982030",[3878],[],{"_212":3879,"_24":233,"_234":3880,"_236":3881},"a0a3bf5f4ae7",[],"“I think both things happen at the same time. Agents need less babysitting every few months, which means people will run more of them, on longer tasks, across more machines. The total amount of agent work is going to grow way faster than the number of decisions one person can make in a day.”","30122ab66bfb",[3885],[],{"_212":3886,"_24":233,"_234":3887,"_236":3888},"84d5d492af88",[],"Gosain reflects on what will happen to projects like his in the future: Will they become less necessary as individual agents become more competent and easier to manage? Or more necessary as more teams onboard larger swarms of agents?","68a0e61fad4a",[3892],[],{"_212":3893,"_24":233,"_234":3894,"_236":3895},"b2d7e327e929",[],"The Future of Orchestration (From the Terminal or Otherwise)","526f25bd891d",[3899],[],{"_212":3900,"_24":233,"_234":3901,"_236":3902},"e081e6730529",[1926],"“A test that can't fail is worse than no test at all, because it makes you feel safe. I found three tests that had been green for months and could never go red. Now every bug fix needs a test that I've actually watched fail with the fix removed.”","5ce0d3fc1b3e",[3906],[],{"_212":3907,"_24":233,"_234":3908,"_236":3909},"efeec16cdd16",[],"“And listen to your users. Copy mode, custom keybindings, and the daemon all started as issues people opened. Not everyone will like every choice either. Some people think mouse support in a TUI defeats the whole point, and that's fine, keyboard-first is still the default.”","65082acf5730",[3913],[],{"_212":3914,"_24":233,"_234":3915,"_236":3916},"ae3f1436f63e",[],"“Third, write things down. I've written more than 40 posts on the TUIOS blog, mostly about bugs I got wrong the first time. It's made me a better engineer, and I think people trust a project more when it's open about its mistakes.”","83b109315657",[3920],[],{"_212":3921,"_24":233,"_234":3922,"_236":3923},"89189c5dda2c",[],"“Second, figure out early who owns what. A lot of the hardest bugs came from two parts of the system both thinking they owned the same data. One fun one was a full scrollback costing 232MB per pane, and paying it twice, once in the daemon and once in every client.”","6709a19906a8",[3927,3928,3929,3930,3931,3932,3933],[],{"_212":3952,"_24":233,"_234":3953,"_236":3954},{"_212":3949,"_24":233,"_234":3950,"_236":3951},{"_212":3946,"_24":233,"_234":3947,"_236":3948},{"_212":3943,"_24":233,"_234":3944,"_236":3945},{"_212":3940,"_24":233,"_234":3941,"_236":3942},{"_212":3937,"_24":233,"_234":3938,"_236":3939},{"_212":3934,"_24":233,"_234":3935,"_236":3936},"03ff68661418",[]," with the fix removed.”","71d9a0844a69",[1926],"that I've actually watched fail","e821897e33d0",[]," than no test at all, because it makes you feel safe. I found three tests that had been green for months and could never go red. Now every bug fix needs a test ","9fcb2fef5254",[1926],"worse","89f7daac29b3",[]," is ","c68efdd5d660",[1926],"a test that can't fail","e7a2992b03e7",[],"“The other half of that lesson is that ","2c567aa855f5",[3961,3962,3963],[3958],{"_212":3959,"_24":1893,"_1894":3960},"f924a1482c30","https://github.com/Gaurav-Gosain/tuitest",{"_212":3970,"_24":233,"_234":3971,"_236":3972},{"_212":3967,"_24":233,"_234":3968,"_236":3969},{"_212":3964,"_24":233,"_234":3965,"_236":3966},"807d067e16ed",[],".”","e6b851d76bb4",[3959],"a separate tool for it","1df3f5508e3e",[],"The creator shares many important lessons learned while building the project. “Testing is the big one. Early on I literally wrote in an issue that I was still trying to figure out how to test TUIOS properly, because testing a TUI with real shells inside it is way harder than it looks. Now the tests I trust most start the real binary, attach a real client and read what's on the screen. I even ended up building ","52e66ff87303",[3976],[],{"_212":3977,"_24":233,"_234":3978,"_236":3979},"b09cc780d360",[],"Lessons Learned: Testing, Ownership, Docs, Community","28c61fd9c130",[3983],[],{"_212":3984,"_24":233,"_234":3985,"_236":3986},"f03d323caae1",[],"“For bigger organizations, I'd expect higher-level tools on top that handle goals, ownership and cost, and TUIOS is built to sit underneath those. Pretty much everything it does is available as a command, as JSON, as an event stream and through MCP, so something else can drive it while a person can still jump into any agent whenever they want.”","bd455af80937",[3990],[],{"_212":3991,"_24":233,"_234":3992,"_236":3993},"731f4a1033a5",[],"For long-horizon runs, the creator points out that sessions require a different layer underneath that supports persistent sessions beyond a laptop close. With a significant amount of time and effort, the creator was eventually able to implement a much-requested daemon to restore sessions post-restart into this project, but larger orgs will presumably need more.","1b8cce8046c1",[3997],[],{"_212":3998,"_24":233,"_234":3999,"_236":4000},"9c0a305c08ff",[],"“When an agent gets stuck I want to see exactly what it sees, type into its session, and take over if I need to. That's really natural in a terminal and pretty awkward from a dashboard. It's also where agents already do their work anyway, with the shell, git and tests.”","ffdc999356dc",[4004],[],{"_212":4005,"_24":233,"_234":4006,"_236":4007},"302606d775e7",[],"While the terminal remains popular for individual devs, Gosain isn’t sure it will scale for major projects across an entire organization. “I think the terminal is the right place for agents to run and for you to step in, but not where you'd set goals and budgets for a whole department.\"","f3e6dbfc6664",[4011],[],{"_212":4012,"_24":233,"_234":4013,"_236":4014},"3a4e64e47a2d",[],"The Future of the Terminal: Long Horizon and Org-Wide Projects","3cc74079e800",[4018],[],{"_212":4019,"_24":233,"_234":4020,"_236":4021},"cf20c36daa7e",[],"Gosain adds that untrusted text remains a potential loophole, as agents will read anything from anywhere and potentially interpret what they read as instructions. Text hidden in PR titles, invisible characters in Unicode, and others. “That kind of labelling really needs to become standard everywhere.”","0dfa637b8b71",[4025,4026,4027,4028,4029,4030,4031],[],{"_212":4050,"_24":233,"_234":4051,"_236":4052},{"_212":4047,"_24":233,"_234":4048,"_236":4049},{"_212":4044,"_24":233,"_234":4045,"_236":4046},{"_212":4041,"_24":233,"_234":4042,"_236":4043},{"_212":4038,"_24":233,"_234":4039,"_236":4040},{"_212":4035,"_24":233,"_234":4036,"_236":4037},{"_212":4032,"_24":233,"_234":4033,"_236":4034},"255ac56e2d27",[],". You should be able to take it away while the agent is still running. And you need a clear record of which agent did what, and on whose behalf.”","95994052226c",[1926],"more","47dfd88b7b4f",[]," than it has but never ","1a9fe7c73af4",[1926],"less","860c3089b9db",[],", so an agent can give a helper ","837e89a39a8a",[1926],"down","ec06fc46f793",[],"“Building grants taught me a few things agents need that most systems don't have yet. Access should be tied to a task rather than a person. It should only ever be passed ","cdb499682691",[4056],[],{"_212":4057,"_24":233,"_234":4058,"_236":4059},"7512596d2d33",[],"The creator explains that while his project focused on least privilege rather than full-blown identity management, teams will likely need IAM systems for both human engineers and agents in the future (and major identity vendors are already hard at work building them).","45b541c5899f",[4063,4064,4065],[],{"_212":4072,"_24":233,"_234":4073,"_236":4074},{"_212":4069,"_24":233,"_234":4070,"_236":4071},{"_212":4066,"_24":233,"_234":4067,"_236":4068},"96fa96fae98a",[],". “Every pane in TUIOS has grants that say what the program inside it is allowed to do through TUIOS: Read its own session, type into it, start other agents, answer prompts, or do everything. The daemon checks every single call against them, no matter how it comes in. So if I want a reviewer agent that can look but not touch, I just start it with read-only.”","2083e36a4d74",[1926],"specifically run through the project itself","c439e261144e",[],"Gosain points out that the project has built-in identity management for jobs ","bd5192f93adf",[4078],[],{"_212":4079,"_24":233,"_234":4080,"_236":4081},"bf4caaeb1172",[],"Managing Agents Via Terminal: Identity and Loopholes","df24faceca60",[4085],[],{"_212":4086,"_24":233,"_234":4087,"_236":4088},"29495e2c94ed",[],"The creator also points out that his project has its limits, controlling what an agent can do through its multiplexer, but not everywhere else. “The ‘ask a human’ path has to be quick and painless, otherwise people just turn it off.”","ed32ddd9d6d8",[4092],[],{"_212":4093,"_24":233,"_234":4094,"_236":4095},"a346f352c481",[],"Gosain suggests that agents generally don’t go astray because they’re inherently malicious. “They go wrong because they're confident and fast. We've all seen the stories of an agent wiping a production database because it found a token lying around. That's what I worry about, an agent being completely sure about something it really should have asked about.”","7918883b39e8",[4099,4100,4101],[],{"_212":4108,"_24":233,"_234":4109,"_236":4110},{"_212":4105,"_24":233,"_234":4106,"_236":4107},{"_212":4102,"_24":233,"_234":4103,"_236":4104},"66999ff3e33e",[]," a risky request if you've allowed it to, but it can never approve one. An agent can't give itself more access, and it can't give a helper more access than it has. Even if an agent edits the config file to give itself more, nothing changes until you apply that change yourself from outside TUIOS.”","dbe58bfdd53e",[1926],"deny","9c3818dff01a",[],"“In practice, that means only you can answer an agent's permission prompt. Another agent can ","d4a501c3e7f1",[4114,4115,4116],[],{"_212":4122,"_24":233,"_234":4123,"_236":4124},{"_212":4119,"_24":233,"_234":4120,"_236":4121},{"_212":4117,"_24":233,"_234":4118,"_236":3966},"79b7b12169fb",[],"fb8f26544685",[1926],"something risky","1bd9d2a22783",[],"According to Gosain, learning from agents on the job is valuable, but only within the bounds of limited autonomy. “The rough rule I ended up with is that agents can do the work and agents can say no, but only a person gets to say yes to ","60319847686f",[4128,4129,4130],[],{"_212":4137,"_24":233,"_234":4138,"_236":4139},{"_212":4134,"_24":233,"_234":4135,"_236":4136},{"_212":4131,"_24":233,"_234":4132,"_236":4133},"729256caecec",[],".” The creator explains that fans let you run your own checks on every attempt, compare the results side by side, and diff the attempts against each other. To help users improve on their own processes, the project allows for annotation that can be shared with agents, effectively giving users the final say based on specific evidence pulled from each agentic run.","715fe426e713",[1926],"comparing","ba5da87edf7b",[],"“The part I care about most is ","f65d4a7bf402",[4146,4147,4148,4149,4150],[4143],{"_212":4144,"_24":1893,"_1894":4145},"3ed3816cd07a","https://github.blog/ai-and-ml/github-copilot/what-are-git-worktrees-and-why-should-i-use-them/",{"_212":4163,"_24":233,"_234":4164,"_236":4165},{"_212":4160,"_24":233,"_234":4161,"_236":4162},{"_212":4157,"_24":233,"_234":4158,"_236":4159},{"_212":4154,"_24":233,"_234":4155,"_236":4156},{"_212":4151,"_24":233,"_234":4152,"_236":4153},"20732e20c65c",[],", partly as a learning aid, but mostly to compare approaches. “Different models fail...and succeed in different ways. If I give the same problem to three agents, I'll often get three reasonable approaches, and seeing them side by side helps me think about the problem. Agent time is cheap and mine isn't, so it's a pretty good trade.”","33fc1103c318",[4144],"worktree","d13353fe843a",[]," functionality to parallelize task execution across multiple agents, each in its own ","46ba47a14fde",[1926],"fan","ca141d2fe81c",[],"The creator notes that he specifically built a ","8fc105268f08",[4169],[],{"_212":4170,"_24":233,"_234":4171,"_236":4172},"291bd897e642",[],"Failstates, Agent Reliability, and Learning Job-to-Job","80403ff8cd0a",[4176],[],{"_212":4177,"_24":233,"_234":4178,"_236":4179},"8d2aebf3369f",[1926],"“Pretty much every lab shipped its own ‘which of my agents needs me’ view, and each one only knows about its own agent. That's fine if you only use one tool, but most people I know don't.” - Gaurav Gosain/Creator, TUIOS","3dca538bee60",[4186,4187,4188],[4183],{"_212":4184,"_24":1893,"_1894":4185},"eb4ac0b4ccd2","https://www.cnbc.com/2026/06/16/anthropics-fable-shutdown-is-a-big-moment-for-open-source-ai.html",{"_212":4195,"_24":233,"_234":4196,"_236":4197},{"_212":4192,"_24":233,"_234":4193,"_236":4194},{"_212":4189,"_24":233,"_234":4190,"_236":4191},"db1ed4e125ad",[]," [access to its tools].” Rather than depend on any one vendor’s roadmap, the project was built to work with multiple tools as well as to be directly compatible with tmux through a tmux shim.","af462a24bae4",[4184],"a lab decided to cut off","63e88dfb40d6",[],"The creator clarifies that his goal wasn’t to create a defensible moat around a single, winner-take-all harness that outcompetes all other labs’ harnesses. “[This] is exactly why a neutral layer is useful. We also saw how fast access can change, when ","c330be734330",[4201],[],{"_212":4202,"_24":233,"_234":4203,"_236":4204},"8217051e8152",[],"“Something I noticed this year is that pretty much every lab shipped its own ‘which of my agents needs me’ view, and each one only knows about its own agent. That's fine if you only use one tool, but most people I know don't. TUIOS recognizes 24 agent CLIs right now, and anything it doesn't know still runs fine as a normal terminal program.”","ab4bc0e3ed05",[4208],[],{"_212":4209,"_24":233,"_234":4210,"_236":4211},"b84cc79e55a3",[],"Gosain is frank about why he decided to keep the project as something that sits on top of other harnesses, rather than trying to also build his own coding agent. “The harnesses get better every week, and the teams behind them are a lot bigger than one person working on an open source project on the side. I also just don't think another agent is what people need. Most people have their favorite, and a lot of them use two or three depending on the task.”","db2551a7b932",[4215],[],{"_212":4216,"_24":233,"_234":4217,"_236":4218},"a7aeb5be88b4",[],"Where Terminal-Level Orchestration Fits in the Big Picture","fa331092dd99",[4222,4223,4224,4225,4226],[],{"_212":4239,"_24":233,"_234":4240,"_236":4241},{"_212":4236,"_24":233,"_234":4237,"_236":4238},{"_212":4233,"_24":233,"_234":4234,"_236":4235},{"_212":4230,"_24":233,"_234":4231,"_236":4232},{"_212":4227,"_24":233,"_234":4228,"_236":4229},"6d9a9be4822a",[]," won't, at least not soon. So review doesn't disappear, it shifts from reading every line to looking at evidence, like whether the tests pass or how two attempts compare.”","03e18c2bea33",[1926],"‘Is this actually the right change?’","14d1aa383e6c",[]," will mostly get automated away. ","459f8e217e2c",[1926],"‘Can this agent run this command?’","f60396b582a3",[],"“I also think the bottleneck moves over time. ","cedcc0d17912",[4245],[],{"_212":4246,"_24":233,"_234":4247,"_236":4248},"472e63a58ba6",[],"The creator suggests that his approach is about streamlining the decision process to fewer, better questions with the assumption that the routine stuff can and will run on its own. “And when something does reach you, it comes with enough context that you can decide in a few seconds.”","c8802a545e46",[4252],[],{"_212":4253,"_24":233,"_234":4254,"_236":4255},"c3b40f8ab7f9",[],"“But I also don't think the answer is asking the human about everything. If you get asked a hundred times an hour, you stop reading and just hit yes.” Gosain suggests that major AI labs are trending in the direction of letting classifiers handle routine approvals, which is probably for the best.","ecf4076fb794",[4259],[],{"_212":4260,"_24":233,"_234":4261,"_236":4262},"dde60bb1e02b",[],"The creator suggests that the bottleneck in human-plus-agent systems is still, from his perspective, human attention and decision-making. “I think that's how it should be. I don't want an agent deciding on its own to force push or delete something important.","4ea2d7f44903",[4266],[],{"_212":4267,"_24":233,"_234":4268,"_236":4269},"96ed4d810536",[],"The creator suggests that the project evolved to help human users spend their limited attention span more wisely. “Agents report what state they're in and the ones that need you [get floated] to the top.” The project collects questions and approvals, and agent turns get saved at git checkpoints to make fixes easier, with cheap verification being something to work on in the future.","25ebcace03e7",[4273,4274,4275],[],{"_212":4282,"_24":233,"_234":4283,"_236":4284},{"_212":4279,"_24":233,"_234":4280,"_236":4281},{"_212":4276,"_24":233,"_234":4277,"_236":4278},"a9b7b4c3a230",[],". Spinning up agents is the easy part now. Every lab ships some way to run them in parallel, and some can kick off hundreds of subagents in one go. But producing a hundred results is cheap, and figuring out which ten are actually good is not. If you can't review it, running more of it doesn't really help you.”","c9a437749d53",[1926],"attention","e98c0abcd22c",[],"“What breaks first, at least for me, is ","7b661806cc77",[4288],[],{"_212":4289,"_24":233,"_234":4290,"_236":4291},"1f0ef4e47a71",[],"Gosain suggests that there’s more to any agentic job than simply how many agents are in play. “I don't think there's a linear curve like five, then 10, then 100 [agents being a breaking point]. It depends on the work. Ten agents doing well-defined changes against a good test suite is honestly easier to manage than two agents doing open-ended design work.”","3f31d5748e7c",[4295],[],{"_212":4296,"_24":233,"_234":4297,"_236":4298},"467d878226a1",[],"Where Do Agent-Plus-Human-in-the-Loop Systems Break?","724b27bc87e5",[4302],[],{"_212":4303,"_24":233,"_234":4304,"_236":4305},"8a8f5b8994d7",[],"“TUIOS already owned every one of those panes, so it felt obvious that it should be the one keeping track.” For the creator, giving cues to help agentic power users like himself was becoming increasingly important (though the project still offers an ‘agents-off’ feature for those who just want a better multiplexer).","23ee3f5a58b7",[4309,4310,4311],[],{"_212":4318,"_24":233,"_234":4319,"_236":4320},{"_212":4315,"_24":233,"_234":4316,"_236":4317},{"_212":4312,"_24":233,"_234":4313,"_236":4314},"5382a94334d1",[]," became the problem. “An agent would hit a permission prompt, draw it once, and then just sit there.” Different screens that were in different project states looked identical at a glance, and would sit waiting for a human thumbs up. “It was a pretty silly way to lose time.”","56cdea4cbb7d",[1926],"which agent needed him","01c30f4c4a0b",[],"As Gosain started running several agents in parallel, arranging panes stopped being the problem. Knowing ","c8f8b537e2f0",[4324,4325,4326,4327,4328],[],{"_212":4341,"_24":233,"_234":4342,"_236":4343},{"_212":4338,"_24":233,"_234":4339,"_236":4340},{"_212":4335,"_24":233,"_234":4336,"_236":4337},{"_212":4332,"_24":233,"_234":4333,"_236":4334},{"_212":4329,"_24":233,"_234":4330,"_236":4331},"bcfd2c3fa106",[],"”","b4a8ddc05174",[1926],"which one of them was waiting on me.","f8ecbb456c9b",[]," became the part I actually cared about. I didn't need help arranging my windows anymore, I needed something to tell me ","f55848acbc67",[1926],"that","52a26e91295d",[],"“A multiplexer sees every program you run, its output, its title, the folder it's in, and yet it never really did anything with any of that. Once I started running coding agents in my panes, ","2c0547539626",[4350,4351,4352],[4347],{"_212":4348,"_24":1893,"_1894":4349},"66e4fafd0d2e","https://github.com/tmux/tmux/wiki",{"_212":4359,"_24":233,"_234":4360,"_236":4361},{"_212":4356,"_24":233,"_234":4357,"_236":4358},{"_212":4353,"_24":233,"_234":4354,"_236":4355},"d57416840743",[]," is great, it always felt stuck in a world of memorized key chords and config files.” The creator points out that while the project was initially about all-in-one window management UX, it soon highlighted a deeper need.","c328590f47fd",[4348],"tmux","b55fff7bf397",[],"Gosain notes that the project started from pursuing a better terminal experience. “I live in the terminal for pretty much everything, and while ","3a705bf2cb04",[4365],[],{"_212":4366,"_24":233,"_234":4367,"_236":4368},"49637b6e2cd9",[],"From Multiple Terminals to Multi-Agent","98f9ba5b1089",[4375,4376],[4372],{"_212":4373,"_24":1893,"_1894":4374},"f04212d3515e","https://www.tuios.dev",{"_212":4380,"_24":233,"_234":4381,"_236":4382},{"_212":4377,"_24":233,"_234":4378,"_236":4379},"9b006e84ff8b",[4373,1926],"tuios.dev","51417f32ca16",[1926],"TUIOS helps users orchestrate agents across multiple terminals. Image courtesy ","d4790a7a95ba",{"_1874":4385,"_24":1876},"image-31b56f7b3c41c15c73d7b410e290c2f05c7c5ede-1600x478-jpg","fa72a247de6c",[4392,4393,4394],[4389],{"_212":4390,"_24":1893,"_1894":4391},"3e20d7e69a4e","https://github.com/Gaurav-Gosain/tuios",{"_212":4401,"_24":233,"_234":4402,"_236":4403},{"_212":4398,"_24":233,"_234":4399,"_236":4400},{"_212":4395,"_24":233,"_234":4396,"_236":4397},"d4349a853895",[]," in response. On the surface, TUIOS acts as a multiterminal orchestration tool, but as he explains below, it’s ultimately a way to address one of the trickiest new bottlenecks in software: The limited attention of a human user.","7ed88089c064",[4390],"TUIOS","50d2ad72640b",[],"But what happens when you start utilizing so many agents at once, and have to have so many terminal windows open that it becomes difficult to keep track of everything? Open-source creator Gaurav Gosain noticed this limitation and began building his project ","5dad60941206",[4410,4411,4412],[4407],{"_212":4408,"_24":1893,"_1894":4409},"52e334b19f82","https://en.wikipedia.org/wiki/Terminal_multiplexer",{"_212":4419,"_24":233,"_234":4420,"_236":4421},{"_212":4416,"_24":233,"_234":4417,"_236":4418},{"_212":4413,"_24":233,"_234":4414,"_236":4415},"932af58ea732",[]," to give each agent their own separate pane that segregates and preserves the work of each across sessions.","ae81e2f7bd1b",[4408],"terminal multiplexers","fcf5bef05f52",[],"With the increasing popularity of command-line interface (CLI) building for agentic projects, terminal interfaces are seeing something of a rebirth in software development. However, as more developers utilize agentic tools, and increasingly, multiple agents at once, they’re turning to ","eef26fcaf517",[4425],[],{"_212":4426,"_24":233,"_234":4427,"_236":4428},"24ae57392a2a",[],"Do Windows Matter Less Than Limited Human Attention?",{"_1976":4462,"_96":4463,"_1949":4464,"_1980":1968,"_4435":4436,"_51":4465,"_14":4466,"_63":4467,"_1987":25},{"_1976":4448,"_96":4449,"_1949":4450,"_1980":1966,"_4435":4436,"_51":4451,"_14":4452,"_63":4453,"_1987":25},{"_1976":4432,"_96":4433,"_1949":4434,"_1980":1964,"_4435":4436,"_51":4437,"_14":4438,"_63":4439,"_1987":25},[4440],751.461333,"https://image.mux.com/WUC7YYclMdxXDY02rzsYF5AyLHBqXYEHf/thumbnail.png","poster","mux","2026-10-26T15:20:00.000Z","persistent-memory-and-what-breaks-in-production","Persistent Memory + What Breaks in Production",{"_212":4441,"_24":214,"_215":4442,"_217":4443,"_219":220},"exmem0",[4444],[],{"_212":4445,"_24":233,"_234":4446,"_236":4447},"exmem0s",[],"Sarah Wooders, CTO of Letta and co-author of MemGPT, on how her team works with long-running agents: persistent memory, continual learning in token space, what breaks when context rots and memory gets polluted, and how to measure what an agent actually learned.",[4454],859.0021,"https://image.mux.com/ZfDsqm02Bjw4U1qiGt2XcKYLXgQo4sdL02/thumbnail.png","2026-10-26T15:25:00.000Z","safety-for-autonomous-organizations","Safety for Autonomous Organizations",{"_212":4455,"_24":214,"_215":4456,"_217":4457,"_219":220},"exsafe0",[4458],[],{"_212":4459,"_24":233,"_234":4460,"_236":4461},"exsafe0s",[],"Axel Backlund, cofounder of Andon Labs and creator of Vending-Bench, shares what agents running vending machines, a market, a cafe, and a radio station reveal about long-horizon autonomy, and why models that can tell when they are in a simulation make real-world deployments essential to safety work.",[4468],1381.3823,"https://image.mux.com/v1FWaHcoYP7EpNyiNbxkyvFqG02JeSzx7/thumbnail.png","2026-10-26T15:30:00.000Z","performance-engineering-for-agents-lessons-from-glean","Performance Engineering for Agents: Lessons from Glean",{"_212":4469,"_24":214,"_215":4470,"_217":4471,"_219":220},"exglean0",[4472],[],{"_212":4473,"_24":233,"_234":4474,"_236":4475},"exglean0s",[],"Tony Gentilcore, cofounder of Glean and a veteran of Chrome's Speed Team, brings a performance-first lens to long-horizon agents in a fireside with Chad Metcalf: reading traces, token budgets that change agent behavior, what work belongs in the background, and measuring spend per unit of throughput.",{"_212":4477,"_24":214,"_215":4478,"_217":4479,"_219":220},"exwes0",[4480],[],{"_212":4481,"_24":233,"_234":4482,"_236":71},"exwes0s",[],"ua","Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)","hideBanner",{"_4505":4506,"_4507":4508,"_4509":4510,"_4511":-5},"ENV",{"_4489":4490,"_4491":4492,"_4493":4494,"_4495":4496,"_4497":4498,"_4499":4500,"_4501":4502,"_4503":4490,"_4504":4490},"SANITY_DATASET","production","ALGOLIA_INDEX","heavybit-library","ALGOLIA_APPLICATION_ID","0QCKBC1FMK","ALGOLIA_SITEWIDE_INDEX","heavybit-sitewide","ALGOLIA_JOBS_INDEX","heavybit-jobs","MUX_ENV_KEY","dRI98bL8ze03gYDL6-Ogle2Js","ALGOLIA_SEARCH_API_KEY","7733fdd199ec7af7ee2923ec39c3f036","MODE","VITE_CONTEXT","footer",{"_4623":4624,"_4625":4626,"_4627":4628},"globalConfig",{"_72":4620},"header",{"_4512":4513,"_4514":4515},"skills","announcementBar",{"_4612":4613,"_4614":-5,"_4615":31,"_4616":4617,"_236":4618,"_90":4619},"headerNavLinks",[4516,4517,4518,4519,4520],{"_212":4593,"_4522":4594,"_4523":4595,"_1893":4596,"_4526":12,"_63":4597},{"_212":4578,"_4522":4579,"_4523":4580,"_1893":4581,"_4526":12,"_63":4582},{"_212":4560,"_4522":4561,"_4523":4562,"_1893":4563,"_4526":12,"_63":4564},{"_212":4528,"_4522":4529,"_4523":4530,"_1893":4531,"_4526":12,"_63":4532},{"_212":4521,"_4522":-5,"_4523":4524,"_1893":4525,"_4526":31,"_63":4527},"3702498a502c","dropdown","label","Contact","/press","mobileOnly","Navigate to contact page","4bec0dd665d4",[4533,4534,4535,4536,4537,4538],"Library","/library","Navigate to library page",{"_212":4556,"_4523":4557,"_1893":4558,"_63":4559},{"_212":4552,"_4523":4553,"_1893":4554,"_63":4555},{"_212":4548,"_4523":4549,"_1893":4550,"_63":4551},{"_212":4545,"_4523":4546,"_1893":4547,"_63":4546},{"_212":4542,"_4523":4543,"_1893":4544,"_63":4543},{"_212":4539,"_4523":4540,"_1893":4541,"_63":4540},"66773779036c","DevToolsDigest","/devtoolsdigest","2ead3bd14715","Reports","/reports","ac06751f0372","Collections","/library/collections","1f3fb936f3b9","Podcast Network","/library/podcasts","Navigate to podcasts page","6526c69f5260","Videos","/library?filter=videos","Watch educational videos in Heavybit's Library","63e48d609548","Articles","/library?filter=articles","Read educational articles in Heavybit's Library","716b76469486",[4565,4566,4567],"Community","/community","Navigate to community page",{"_212":4575,"_4523":4576,"_1893":4577,"_63":4576},{"_212":4572,"_4523":4573,"_1893":4574,"_63":4573},{"_212":4568,"_4523":4569,"_1893":4570,"_63":4571},"23077fca0e2f","DevGuild","/devguild","Navigate to devguild page","1fb7c9c465d0","Advisors","/community/advisors","3a177efc4649","Events","/events","90b68b861c06",[4583,4584],"Portfolio","/portfolio","Navigate to portfolio page",{"_212":4589,"_4523":4590,"_1893":4591,"_63":4592},{"_212":4585,"_4523":4586,"_1893":4587,"_63":4588},"e50fa43ad2e7","Jobs","/jobs","Portfolio Company Jobs","d481fcbbd3af","Spotlights","/portfolio/spotlights","Navigate to spotlights page","a8d641babed2",[4598,4599,4600],"About","/about","Navigate to about page",{"_212":4608,"_4523":4609,"_1893":4610,"_63":4611},{"_212":4604,"_4523":4605,"_1893":4606,"_63":4607},{"_212":4601,"_4523":4602,"_1893":4525,"_63":4603},"84e68f9f5c33","News","Navigate to press page","3d279775d73b","FAQ","/faq","Navigate to FAQ page","964627410473","Team","/team","Navigate to team page","buttonText","RSVP","end","show","start","2026-10-20T21:46:59.960Z","🐺 [Oct 28] GitBlame: Werewolf Happy Hour @GitHub Universe","https://luma.com/gitblame-werewolf",{"_24":1949,"_74":4621},{"_1874":4622,"_24":1876},"image-32fdfae4fd13b8f9a3c5bf6b13a635068cb0ecbb-1332x529-png","emailSubscription",{"_1819":4694,"_4695":4696},"footerColumns",[4660,4661,4662],"socialMedia",[4629,4630,4631,4632,4633],{"_212":4655,"_4635":4656,"_90":4657},{"_212":4650,"_4635":4651,"_90":4652},{"_212":4645,"_4635":4646,"_90":4647},{"_212":4640,"_4635":4641,"_90":4642},{"_212":4634,"_4635":4636,"_90":4637},"639729864c2d","icon",{"_24":1872,"_74":4638},"https://www.tiktok.com/@heavybitindustries",{"_1874":4639,"_24":1876},"image-48ac9473e30341c4bee9f6634b77b6f8f1dd6b2f-849x980-webp","a100f6448036",{"_24":1872,"_74":4643},"https://www.instagram.com/heavybitindustries/",{"_1874":4644,"_24":1876},"image-b38b7ee9ccd05f61ca43d5b29fe7058f822f87eb-512x512-webp","fae1cc0df5c6",{"_24":1872,"_74":4648},"https://www.youtube.com/c/heavybit",{"_1874":4649,"_24":1876},"image-6c548d2e815d1fb3b3077c9b8be1ec251e1bc418-512x512-png","5348b028f0d41c3dac62b3dd2b5531ee",{"_24":1872,"_74":4653},"https://www.linkedin.com/company/heavybit",{"_1874":4654,"_24":1876},"image-01bd92eb06873bdc30b1cb5f9c78b1ef5a9559d4-512x512-png","ee025c96397c",{"_24":1872,"_74":4658},"https://twitter.com/heavybit",{"_1874":4659,"_24":1876},"image-9d618da0948299428b29e5d499af3fa5369a24d2-20x20-svg",{"_212":4686,"_4664":4595,"_1969":4687},{"_212":4674,"_4664":4562,"_1969":4675},{"_212":4663,"_4664":4530,"_1969":4665},"ea618aa34610","heading",[4666,4667,4668],{"_212":4671,"_4523":4672,"_1893":4531,"_63":4673},{"_212":4670,"_4523":4549,"_1893":4550,"_63":4551},{"_212":4669,"_4523":4540,"_1893":4541,"_63":4540},"96a97ceba3a2","e1673433290e","c82a64edc4dd","Blog","Navigate to browse on library page","a584a6acf07b",[4676,4677,4678,4679],{"_212":4685,"_4523":4576,"_1893":4577,"_63":4576},{"_212":4684,"_4523":4569,"_1893":4570,"_63":4571},{"_212":4683,"_4523":4586,"_1893":4587,"_63":4588},{"_212":4680,"_4523":4681,"_1893":4682,"_63":4681},"e22bfa6b2330","Inclusion","/community/diversity-and-inclusion","feb038ce46a1","e2d5641ef7de","4358699f4ca4","30a87ef6e7c1",[4688,4689,4690],{"_212":4693,"_4523":4580,"_1893":4581,"_63":4582},{"_212":4692,"_4523":4609,"_1893":4610,"_63":4611},{"_212":4691,"_4523":4602,"_1893":4525,"_63":4603},"398a9b50e214","42666d0b3987","3dfef1283ab5","Unlock success from technical pioneers in software infrastructure. We send the best strategies, tools, and industry updates directly to your inbox.","targetURL","https://share.hsforms.com/1_KYaR4lAQm-C8_9H7r4MIwqdoyu"]

The Mythical Agent-Month

Speakers