
Ep. #16, The Red Queen's Race with Jennifer Riggins
On episode 16 of Third Loop, the Progressive Delivery team welcomes back Jennifer Riggins to ask whether AI is bringing teams closer to customers or just helping them ship more code. They explore why faster development creates new coordination problems, where reliable automation beats AI improvisation, and why customer feedback still matters more than output.
Jennifer Riggins is an independent freelance technology journalist and event and podcast host whose work appears in publications including The New Stack and LeadDev. She specializes in developer experience, developer productivity, platform engineering, and the impact of AI on software teams, connecting technology with business outcomes and workplace culture.
- Progressive Delivery: Build the Right Thing for the Right People at the Right Time
- exe.dev
- The Third Loop Ep. 13: Ten Times Faster at What? with Jennifer Riggins
- The Third Loop Ep. 11: Durable Systems, Disposable Code with Chad Fowler
- Steve Yegge: Welcome to Gas Town
- Spotify Engineering: Beyond Winning – Spotify’s Experiments with Learning Framework
- Confidence by Spotify
- Monki Gras
- Jessica Rose
- Red Queen’s Race
- Pace Layering (Stewart Brand)
- Heroku
- Salesforce
- VMware
- OpenAI
- ChatGPT
- Gemini
- Sublime Text
transcript
Customers, Canaries, and the Promise of AI
Heidi Waterhouse: When we're thinking about progressive delivery and how it fits with the new way that we're working, what do you think we're missing? What do you think we need to be talking about more?
Jennifer Riggins: I think it's a boring answer, but comes back to the business, like the customers. What do the customers want? And that can be in platform engineering or it can be with your external users.
Progressive delivery is very powerful. Marketing has been doing it for decades now for a lot of it. First and foremost, it's about not breaking things and making sure you can roll back if you do. So all the security and quality stuff has to be there.
But then the interesting stuff that AI should help you do is test things really quickly, but we're not hearing about that anymore. Have users that maybe are your canary. I feel like guinea pigs maybe, often guinea pigs more than canaries in this situation, where you are treating them like the first users. But they get to try all the weird shit and give you feedback.
Heidi: The consumer board capybara.
Jennifer: Consumer board capybara. Thank you. Yeah, just bring teams closer. That was the promise of AI, that teams would get closer to their users, but I don't think that's happening.
Heidi: No, I think we've really detached them because that chat interface is an intermediary. It's sort of like we're now talking through this screen all the time.
Jennifer: Not good.
Heidi: Yeah.
Jennifer: But that was the promise, wasn't it?
Heidi: I don't think the promise was ever to consumers.
I think consumers have had chatbots for a while, and the promise was actually that you would have to talk to your consumers less because somebody would intermediate that with a chatbot and you wouldn't even have to take bug reports. Automate that.
So who was promising AI to who? You were paying attention. I was steadfastly ignoring it in '23, '24.
Adam Zimman: I think that the hype was that AI was going to be the thing for everyone and everything. And that still continues to be the hype, I would argue.
I think that even in the early days of OpenAI, there was the hype of, "Oh, no, we will create this glorious utopia where humans can spend their day in deep navel gazing and contemplation while the robots around us will do all of the things to meet our needs."
I think that the idea that robots would take jobs was the flip side of that articulation, right? I think that it was the parts that, when brought up, were very quickly brushed under the rug were, "What happens to all the people who lose their jobs because of AI?" "Oh yeah, don't worry about them, they'll be fine," as is so often the case with the introduction of new technology.
Let's be fair, this is not unique to AI. But if you want to get right down to it, you think that was the value proposition from a top-down perspective of, "Oh, you can have robots or computers do all the things for you. You don't need to navigate the messy middle and the difficult part of humans."
I think from a bottoms-up perspective, it was, "Well, now you can finally get your answers faster or have your own personal assistant to do all the things for you." And the idea of convenience was definitely a huge selling point there.
AI vs. Automation
Heidi: So what do we think AI does that isn't automation? Kim and I were talking about this. Is there anything AI does that isn't automation?
Jennifer: Well, if it's not automation, it's probably not AI. And if it could have been solved with automation, we probably shouldn't have used AI. Environmentally risk-worthy things like that.
AI is supposed to do the work that— If you can automate something, still use automated scripts and deterministic.
But if you are trying to work out a problem like finding a bug that is nuanced, a unique problem that is not something a human could do on their own easily, finding a complex bug in a complex system and suggest an interesting or unique solution, that's where AI is interesting.
Or even if you want it to act on that solution. People, when you're asking it to act on that solution, is that really AI or is at that point a script you're approving for automation?
Adam: Well, I think that this is also consistent with what I've seen of teams that are using AI effectively with regards to building and running systems, is that the expectation is that the AI is actually being used to build scripts or being used to build automation in a way that is creating consistent and predictable outcomes.
And as opposed to just the YOLO approach of, "Oh, well, I'm not going to worry about it. I'm just going to say, oh, figure it out, and do it as a one-off, and then I don't care whether or not you can do it again."
Jennifer: It's wild because since when did we want non-deterministic outcomes in a business, right? That doesn't seem to be the goal of anything.
If you're experimenting, like we were talking about Spotify earlier, they run over 10,000 experiments a year. That's why everybody has these different experiences and all. And they only run it for a small group of their, say, 3 million users, they'll run it on at a time. You can really do that if you have that many users.
But that's still very controlled. They have the guardrails in place and they know what they're doing and they're trying to find out an outcome. So they're running an experiment.
But with a lot of our software, we want things to be deterministic still. We are not running experiments on our entire population. Where if we are, it's the bubble we're living in right now and we're all in an experiment and we're screwed.
Small Teams, Big-Team Problems
Kim Harrison: One of the things that I was thinking about when Heidi and I were talking about this is there are things that we can automate and then there are people problems that still require people. And I think the last time we chatted, you actually talked about what an organization should look like, what the internal structure should reflect if you're going to start inviting AI into workflows. But are people trying to chip away at people problems with tools that really can't solve that?
Heidi: Are you saying that people would use tools over process in humans?
Jennifer: Are you saying that people would use AI in recruitment processes for the last decade?
Adam: So I've got a slightly different angle on this that I think I'm starting to hear and see personally in the small startup that I'm at now. And teams that are using AI rather significantly are realizing that they are able to have successfully increasing their output.
And I would say that our organization, which has about nine developers, that probably has a equivalent output from a few years ago of a team of around 40, conservatively, just their senior engineers. They're able to do a lot. We're starting to run into the same challenges that you would have with a 40-person engineering organization with regards to tracking and awareness of the scope of what we're working on.
And you think about this just simply in the context of, it used to be that if you had seven people or eight people on an engineering team, you could have a weekly sync on what everyone was working on and it would take maybe 15-30 minutes to be able to go through what projects folks were working on and what you need to think about in terms of scope from a architecture or design perspective for future projects. There's no way that we can cover the amount of what everyone is working on in that kind of short weekly meeting.
Jennifer: You're saying a daily stand-up doesn't work?
Adam: Especially if you want people to actually have context, right? It's one thing for you to just have a checklist of, "These are the things that I'm sending my agents off to do today or even this week," if you can conceive of what that's going to be for a week, because that's even hard.
But the reality is that the scope has expanded along with this notion of output. And all of a sudden, the coordination of even a small team is becoming that much more challenging. And having people even have awareness of the breadth of the system.
Now, granted, there's definitely agents that can help in that regard. You can ask, you can interrogate a system from an agent's perspective in a way that you couldn't a few years ago. But still, it presents people challenges, right? It's the coordination effort.
It's the mental effort of being able to hold that in your brain and being able to be effective. And you get back to challenges of duplication of effort or you get back to challenges of people diverging in terms of what we talk about in the book around alignment of, "What are the priorities?"
All of a sudden, a lot of those challenges arise much sooner with much smaller teams than they were in the past.
Does a Nine-Person Team Need a Platform?
Jennifer: We used to say that a company of about 50 was when you started to need a platform, an internal developer platform. Like 50 engineers, you lose that sense. Is that becoming reduced because of the both output? Though I don't love that word. It sounds a little like human factories.
And because of the agent— I hate the word teammates and that too, but would a platform help that? Especially because it sounds like you're doing a lot of duplicate work for usability because you don't all need to create agents all day long. You could reuse other agents. Would a platform work even though you only have nine people?
Adam: So in this particular instance, I work at exe.dev these days. And so we build a platform and we are all using it. But that doesn't mean there aren't— Because we're building a platform, a lot of the work we're doing is on, how do we want to expand it? How do we want to offer new functionality? How do we want to try new things?
And so a lot of the developers do have their own variants or their own flavors that parts of it get merged back in, parts of it don't, parts of it go totally off the rails and get scrapped.
But I think that this is the thing, is that I've seen this even in organizations that aren't building platforms, where it depends, right? There's some organizations that are trying to say, "Hey, we're going to standardize on our agent harness. We're going to have a platform team that builds that. And then the engineers can go, other developers can go and use that."
But there's also people who are like— This is engineers. It's like kids in a candy store, right? They're like, "Oh, well, I kind of want the experience of building my own. I read Steve Yegge's post and I want Gas Town," right?
Heidi: No!
Adam: And I think that this is where I'm seeing some really interesting things happening with the way that individual engineers are choosing to explain their own productivity and the tools that they need to be productive.
Jennifer: First of all, I would hope in a nine-person team, people don't need to explain their productivity because there would be a level of trust that everyone trusts that people are being productive.
And productive maybe is no longer the goal if the price of code is zero.
Business value is always the real number that we've cared about, like profit, retention, things like that. So it sounds like to me, and I'm not a tech therapist, but it sounds to me that you as a platform engineering company have the problem of having all your platform engineers that think they know what's best in a platform. But are you talking to your customers enough?
Adam: Yeah. And I think that for us, I think it is one of the things where I feel like we're managing it quite well. And I think that we're being pretty good about, number one, talking to customers on a pretty regular basis.
And this is the other part that's kind of weird with such a small team of senior developers, that we've chosen to align ourselves with developers that actually like feedback from customers and like talking with customers. So every single developer on the team is having regular conversations with developers and engaging in that level, which I think helps.
That said, it's still a lot for us to be able to manage the scope. And it's far more challenging than I've ever experienced at such a small team, because we're doing so much more.
Can We vs. Should We
Jennifer: Should you be? Is not everyone working nine to 12 hours a day now because they can?
Adam: No, well—
Jennifer: Wasn't the dream of AI is backing off work and just working on the really important stuff?
Adam: And I think that this is something that I've noticed that I'm happy about with our org, is that we are actually backing off. And it is not what you hear about from a lot of organizations that are having these crazy benders of, "I'm going to work with AI for 15 hours a day, eight days a week."
But I think it's still really impressive in terms of just the amount of work that is able to be done. And we're managing alignment pretty well, but the pace is significantly faster than you'd expect for such a small team.
Heidi: Yeah, but I think what Jennifer is saying, and I think this is super interesting, is should we be doing that much?
Adam: I think you're right in the sense of, we do have regular conversations of the difference between can we and should we. Just because you can doesn't mean you should.
And I think that this is also a function of one of the challenges of being, even though we're not doing frontier model development, we are engaged in trying to build a platform that is useful for agents. And that world is changing so rapidly.
So a lot of the things that we're doing is making an attempt to keep up with trends and paces of the messy middle, where we don't know where things are going to land. And so a lot of it is still in the aspect of throw stuff at the wall and see what sticks. It's just the pace of that is significantly faster than it used to be.
Heidi: I keep thinking about the Red Queen's race. So this is from Alice in Wonderland. And the Red Queen grabs Alice's hand and runs and runs. And Alice is like, "Why are we going so fast?" And the Queen says, "Oh, we have to run this fast just to stay where we are."
And I feel like as an industry, that is very much what we're doing right now: we have to run very fast to stay where we are.
But do we? Or could we be slower and more mindful like that? The messy middle that you're talking about, could we say what we're doing is building something more timeless? Do we know enough to build something timeless?
Adam: From what I've seen, I'm not convinced we know enough. And the reason that I say that is that I feel like as the models improve and as people are using them in anger, I see more and more that the primitives are changing.
This goes back to what we talked about with Chad Fowler, right? It's the pace layering. I think that there's definitely some layers that, yes, we can be thoughtful and mindful of. But there are a number of layers right now that are really being disrupted.
And I think that that is changing the way that we're trying to work as teams. I'm seeing this not just from an engineering perspective, but from other roles in organizations as well, right? From sales to marketing, whatever your role used to be, the tools that are available to you and the way that you're able to get at the same outcomes that you wanted five, 10 years ago is changing.
Heidi: But to Jennifer's point, are we sure that we want that much more outcome?
Jennifer: Yeah, our end users want more outcome, I think. And Jessica Rose is the first person who ever said, years ago at a Monki Gras, a decade or more ago, but I've not aged a day.
And she said that the tech industry is the only industry outside of drug dealing that we talk about users.
And we as the creators have now trained the AI to get us addicted to. And it could be capitalistic pace and all. There's a lot going on in there, and I don't know if it's just addiction or if it's also just fear.
Kim: They had practice. Okay, they've been practicing this with social media. They know what they're doing.
Jennifer: Yeah, and now it's the creators of the tech and everyone else too, but it may be both addiction to creation and this dopamine hit, but it could also be plain old fear, because if we slow down, we're going to be cut because of stupid tokenmaxxing and other ideas.
Kim: It's a gold rush.
Heidi: So if it's a gold rush, who's selling the shovels?
Jennifer: Who said it's gold?
Adam: Ouch!
Jennifer: There was a lot of fool's gold in the gold rush. Not a lot of people made money on the gold rush.
Kim: Hey, that's real.
Adam: And there were some people who made money on the fool's gold too.
Staying Close to Customers
Heidi: So Jennifer, what do you think teams should focus on then? If we're building platforms, what should we be thinking about?
Jennifer: The customers. I don't know what periodically meant, Adam, when you said, "Well, we're talking to our customers fairly periodically or fairly often." I don't know what that means.
Adam: Yeah, for us, we're talking with customers on a weekly basis. And customers, multiple. But we're also like, small team, right? We're a small team, early stage startup type product.
So if we're doing it right, we're talking with customers regularly. And I feel like that is right now, from what I've been seeing and doing, that is happening.
But I think that that's definitely not all teams that are doing that. And I think that the interesting part that I've seen from a customer perspective is we are definitely at a stage where the needs of customers, or what they articulate as their needs, is also changing very quickly. And part of that is that they're trying to figure out the products that they're building.
I think this is also very different from the world I was in a year ago, working on Heroku at Salesforce, where we didn't speak to customers much at all. And the needs of the customers that we did talk to was very slow paced. It was larger organizations that had their own cadence in terms of change that was very different.
Jennifer: They also have insanely long enterprise SaaS contracts, so there's no need to talk to customers. But it is interesting because this was the big promise of AI, was that the people that are successful with AI are going to be closer to the business and closer to the customers. And I'm not seeing it in general.
It may be happening at a highly iterative experimental startup level where you can get prototypes in front of customers faster, and that's great. But a mix of risk appetite and people don't want to talk to customers has this pushback that it's not happening, I think.
Adam: Yeah. No, and I think that this is partly where I am where I am, because the team that is actually aligned with my personal belief that you don't know what you need to build unless you're actually talking to customers.
But I hear you, and I definitely, being in San Francisco and having gone to some of the AI meetups and conversations, I see consistent with what you're seeing by and large, right? This is something where there is a lot of, "Well, I know what needs to be built" kind of mentality that, in fairness, I remember back when I was at VMware, there was a lot of engineering mentality that shared that, where, "Don't tell me what the customer wants, I already know."
Jennifer: Yeah, and it's never been a way for success unless you have one of those great enterprise contracts and a good lawyer. And they can't drop you.
The Work vs. the Job
Adam: How are you seeing the changes taking place between what is historically tech buyers, whether that's enterprise or even just a tech or IT teams in normal businesses, non-tech focused businesses, versus the consumer buyers or the consumer users, that is starting to get a little bit more muddled and mixed with some of the conversations around AI? Is that something that you've been looking at? Is that something that you have seen significant changes in? Or do you feel like it's just fake news?
Jennifer: First of all, it serves to all of us because everyone Googles. So everyone is interacting and becoming onboarded with Gemini because the second you ask actually a follow-up question, it opens into a Gemini situation. And people are doing that in ChatGPT, etc.
People are writing less and they think they're more efficient and effective at communicating this way. So it is happening across the board. I mean, a certain age range and things like that.
I'm terrified for students that are just living on this. It's terrifying. But I'm just not cool like that. I don't do that.
So I just think the best thing to teach is don't use AI for unique value proposition. I don't use it for writing.
Heidi: I had a great session yesterday where I used it to debug a bash script in my Sublime with a linter. And it was actually super useful because I didn't feel bad asking stupid questions, which I do. I do. I should be better about it.
But I wouldn't use it to write because one of them is my work and one of them is my job.
Like getting my blog publishing, that's just my job. I don't care. I'll use AI for that. But the work of writing, I wouldn't use AI for. And I think there's a lot of people who that's reversed for.
Jennifer: How boring. What a generic mid-world we're heading towards, and we're okay with that.
Heidi: I remember when, what was it, DeepMind? The weird, trippy visual one came out. And people were just making all sorts of very strange art with it.
And I'm like, "This is cool. This is not trying to fake something. This is causing glitches that tickle our brain." And I thought that was much more interesting than everything that's realistic.
Consent and Accountability
Jennifer: Versus using it that— I just saw a statistic today that 70 plus percentage of parliamentarians in Europe that are women have had deepfake porn against them, and a couple men. Porn for men, pat on the back. Porn for women, and sex work is work, but it just can destroy their lives.
Heidi: Yeah. And it's so aggressive.
So here's the thing that I keep coming back to in my struggles with AI is consent. And we didn't ask permission to ingest the world.
Adam: Well, it's consent and accountability, I think, which are the two ends of the work stream, right?
Heidi: There was no consent up front.
Adam: And any harm that's being done, there's no accountability currently for.
Heidi: Yeah. And I think that's an interesting problem that we are really unprepared to struggle with.
Adam: Well, I think that not only unprepared to struggle with, I think we're ill-equipped from a societal structure to be able to actually do anything about. I think we've talked about this before on the show, is that government is not set up to move at the pace that technology is moving at.
And I think that the expectation or the mantra, if you will, from so many of the companies that are doing harm is, "Well, it's not my responsibility to not do harm. It's the responsibility of the government to regulate me and they're not doing it." So there's this complete abdication of responsibility.
Heidi: No one said I couldn't.
Adam: Right.
Jennifer: In fact, the creators of this demon AI right now are even scared and saying, "Please regulate me." Which means they should self-regulate, but we're clearly not good at that.
Content from the Library
Third Loop Ep. #13, Ten Times Faster at What? with Jennifer Riggins
On episode 13 of Third Loop, the Progressive Delivery team speaks with Jennifer Riggins about why faster code doesn’t automatically mean better software. They explore how AI exposes organizational bottlenecks, why psychological safety matters, and what Spotify’s growing feature set reveals about user value. Jennifer makes the case for smaller experiments, stronger feedback loops, and keeping people at the center of delivery.
Third Loop Ep. #12, Shipping at the Speed of Understanding with Steph Hippo
On episode 12 of Third Loop, the Progressive Delivery team speaks with Steph Hippo, Platform Engineering Director at Honeycomb, about what happens when AI accelerates development beyond the pace humans can absorb. They explore safe defaults, evolving code reviews, and the feedback loops that help teams understand what they ship. Steph makes the case for reinvesting saved time in better decisions, software quality, and human connection.
The Future of Software Documentation in the Age of AI
Technical writing expert Heidi Waterhouse explains why documentation needs to evolve to accommodate the way AI is changing software development.



