Ep. #36, Solarpunk with Christine Spang

Rachel Chalmers: Today, it is awesome to have Christine Spang on the podcast, actually a repeat guest because you were here with the AI DevGuild as well.

Spang is co-founder and CTO of Nylas, a communications API platform that lets companies and developers unlock the value hidden in communications data to make it incredibly easy for developers to build software that helps the world communicate and connect better.

Before founding Nylas, Spang worked at Ksplice and Oracle as a principal dev, focusing on backend systems.

She went to MIT, and in addition to the time she spends building internet infrastructure, she speaks at major conferences around the world and lives in the wonderful city of Oakland, California.

Spang, thank you so much for coming on the show.

Christine Spang: It's great to be here Rachel, always fun to hang.

Rachel: Let's start with your origin story. Where did Nylas come from? What inspired it?

Christine: Yeah, for sure. It's been a long time, so I'll try to keep this quick.

But the short version is that, so I went to university at MIT, and the reason I ended up at MIT was because I was really into open-source software, and I contributed to Linux when I was in high school, and really just was my dream school.

And so I ended up going to MIT. And there, when I was at MIT, I met some folks who were sort of at the intersection between open source and the startup community.

They were building a software company that basically commercialized a service that did things for Linux.

So, I ended up working for them for a little bit and then I actually sort of went through like this accelerated tiny startup journey with them where I worked with them, they sold the company to Oracle, and kind of just got the startup bug through that.

Like, I definitely had no idea what a startup was before I joined Ksplice, and I really thought of myself as like this hardcore technology person, and I'm like, "I'm going to like build distributed operating systems or something crazy like that."

But got the startup bug, decided that like I really was passionate about kind of the applications of technology and not so much sort of sitting in a lab and figuring out sort of academic-style innovations.

And basically, what happened is I met another friend who was sort really struggling with this problem of basically building a customized email app.

And that's where the idea came from that led to Nylas, basically seeing that my friend was trying to build this app that did things with the data inside of email, but he just spent like all the time just dealing with email plumbing, like how to extract the message, display the message, connect to the protocol.

At the time, really, the only way that you could connect to an email mailbox was using this protocol called IMAP.

And I don't bore you with too many details, but suffice to say that a modern web developer would not find IMAP to be an intuitive way to work with data.

And so when basically, my retention period was up at Oracle was trying to figure out what else to do and decided to start a company.

And so I moved from Boston to the Bay. It was kind of just like a time in my life where I felt really open to doing something that might not go anywhere.

So I started Nylas, and the super short version is like I spent four years building like an email client with some infrastructure behind it. The email client ended up being a bad business overall, like just in terms of monetization, but the infrastructure part ended up being really valuable.

So we built all these APIs that basically abstracted away the complexity of connecting to email mailboxes, and then eventually calendar and address books as well, and came away with this unified set of REST API infrastructure for manipulating and working with that valuable productivity data.

And when the client didn't work out, we spun it off. We had actually enough people that were using the infrastructure that we just doubled down on that as a business.

And that was many years ago. And that's basically the same thing that we're doing today.

Obviously, it's evolved and matured, but that's really kind of the root.

Rachel: I should confess to the audience that Nylas has been an incredible learning experience for me 'cause I had the opportunity to invest very early when it was an email client business.

And I wasn't totally sold on the email client, so I didn't sell it to the partners at my firm as hard as I should have.

And in so doing, I missed out on one of the great infrastructure success stories of the last 10 years. So, it keeps me honest.

Christine: Well, it turns out that evil client was a bad idea, but the people were amazing and there was some good tech there.

Rachel: Yeah, and what the experience taught me was when I meet a founder and I find that founder really compelling, even if the business isn't something that I'm going to fall in love with, the founder is going to do something amazing with it.

So you, even more than I have seen the enterprise software world change so much over the last decade.

And now we've got gen AI in the mix, we've got talking robots. What's hype and what's real?

Christine: Yeah, I think there's like a few different things I wanted to share about just sort of generally how the software ecosystem and business software has evolved over the past decade plus of kind of being a part of that ecosystem.

One is like I know in kind of the consumer world, people talk about consumer products go through these phases of like unbundling and then you do bundling. Software kind of does the same thing.

So, one thing that I've seen a lot is that when sort of a new piece of functionality comes out, it comes out as a point solution. Like, I have a dedicated tool to do this one particular thing.

And one example of that right now is, for example, AI note takers.

You have all these point solutions that you can use alongside whatever tools that you're using, but alongside the introduction of these specific tools, people are building habits, and it's sort of raising the bar for like what people expect from technology in general. And there's kind of this co-trend, I would say, of there being sort of one leader that raises the bar for everybody else.

Like, for example, in the chat world or sort of a business chat, before Slack was a commonly used tool, people had a really terrible experience using chat at work.

People were basically emailing each other back and forth because they didn't have this concept of like a really nice, well-designed, easy-to-use kind of consumer-style tool that you could use in your day-to-day work.

And the bar for user experience in enterprise tools was very poor. There wasn't a lot of choice that people had.

And one thing that has really shown up in a major way over the past decade is that some players raised the bar and then everybody else sort of has to kind of rise to meet that bar.

And so I think like with AI note takers, like with business productivity software, like it's just going to be an expectation in a couple of years that if you are having meetings, that that software is related to whatever sort of workflows you're dealing with on that software, you're just going to have access to the transcripts from the meetings, summaries, the action items, and this sort of like idea that you would like manually take notes for a conversation and have to like input them into another system is just going to completely go away because everybody's learned that this is like awesome, it makes my life way better. And I don't want to do it manually anymore.

So when I'm out there looking for tools, for platforms for things that I can use to run my business, it just becomes like a feature checkbox on like what software I end up deciding to use.

So, I would say that I've seen a lot of this sort of like bundling trend happening, where you start with this like unbundled set of features.

And then people want sort of a seamless experience of things that kind of all work together, and play nice, and like you have the data that's relevant and all the context put together, and that means that like kind of the platforms that are winning today are kind of the most integrated experiences.

Where that's going? We'll see. I think that like in the future, there's a lot of hype and talk about...

Maybe in the future, all of our user interfaces are going to be AI agents, and we're not going to use kind of like SaaS with like point-and-clicky buttons and things like that.

I think that's a ways away, except for limited use cases and probably won't play out exactly the way people are imagining it right now.

It's sort of in the very exploratory stage, but people have big companies are talking about it quite a lot and really taking it seriously as like, oh, we need to like figure out how do we operate in a world where maybe the way things get done is sort of more agentic than the traditional type of kind of SaaS platforms that people have been trained to use over the past 20 years.

But it's something that goes slower than you might expect, especially in terms of like uptake across everything.

And I think for like AI stuff today, like the real value is coming in like very specific areas, where basically the functionality of the tech that we have today meets a specific use case that is okay with kind of the downsides of the current limitations of tech.

Rachel: With Nylas, you're sitting on this enormous amount of really valuable communication data.

Not only the data that people are sending each other, but your first derivative understanding of how people communicate within organizations and what those patterns look like.

AI eats data for breakfast. How do you think about responsibly using that communication data to power AI features?

Christine: Yeah, for sure. This is something where we've had a strong perspective on, even before kind of like the current trends of AI.

And our perspective is that the data that people are connecting to through our system, this belongs to the company that's using it and the end users at the end of the day.

Like, it's not our data. We're an infrastructure provider that's helping you get access to that data.

But we've always had a very strong sort of security stance on that data belonging to the people that are actually the sources of that data and the users of that data.

We have certain AI features in some of our products, but the entire platform, like if you're not using an AI feature, your data's not being used for AI purposes.

Like, that would be something... You need to be transparent when it comes to what your data is being used for.

And I think that's really as just a continuation of what has been happening in general, like, even outside of data.

Like, the EU has strong data protections. California does now, too. And people really care about what's happening with their data and the core sort of foundation of that is transparency on what it's being used for.

And so the vast majority of our platform, like we're not training models with any of that data, it's just being used to provide the service that people are consuming.

And then for like specific AI features, there's a opt-in around that. And if it's being used to to train something, you're going to know.

Rachel: Yeah, you got to earn that trust.

Christine: Yeah, I mean, we support businesses that also need to like provide strong stances to their customers.

So, it's really something that's important to us to be really smart about how we use that data and make sure that we are establishing that level of trust that people expect.

Rachel: So, back to your point about the specific use cases. Some people are skeptical that there are real enterprise use cases for AI at all.

Where do you see the wins, and where do you see the question marks?

We talked about agents probably won't come as quickly or be as powerful, maybe, as we think.

What do you see as the low-hanging fruit?

Christine: Yeah, the low-hanging fruit, things that are already useful today are things like plugging your internal knowledge base, or for us, like we have like, for example, pretty extensive documentation for our product 'cause it's a developer product.

There are tons of providers out there today that you can just sort of plug in pointer to all that documentation and get like a brand-new user interface for people to consume that data.

I think the chat interface for sort of interfacing with the encyclopedia is actually really useful.

People prefer to ask questions and to get answers to those questions than to like look up in the index where do I find in this book the answer to my question.

That's a place where it's just a better user interface.

Rachel: Yeah, not to be too Gen X about it, but we all laughed at Clippy, the helpful paper clip in Windows 95, and now we talk to Clippy all day every day.

Christine: Yeah, that's because Clippy was bad, and now it's good.

Rachel: Yeah.

Christine: So that's one example.

Another example that I've seen a lot of change in like positive uptake is in customer support. So things like doing kind of first-line triage of questions.

Complex things still end up in the hands of a human, but there's a lot of kind of like rote-type work of just like identifying, like: What kind of question is this? Do we have a knowledge-based answer that can really be automated?

That large language model technology has actually made it a lot better, and I think people are actually getting a lot of value from that kind of thing right now.

And there's lots of companies out there that are essentially deploying this technology in the enterprise space today because it is valuable.

So there's kind of two things that come to the top of my mind.

In general, the theme is finding the places where automation can be pushed forward by having a machine that understands some of the context of unstructured communication in a way that wasn't possible before. It's super cool that I can ask a question that maybe is not fully well-formed and the machine just gets the gist of it most of the time.

Rachel: Yeah, it's autocorrect but for your--

Christine: For your brain!

Rachel: How are your customers thinking about AI? Are you having to push this stuff on them, or are they pulling you in?

Christine: More of a pull, I would say. As an infrastructure provider, like I would say that like probably like 1/4 of just new companies that are signing up for our platform right now have like a .ai, really some sort of like AI-type thing in their domain or in kind of their messaging.

Rachel: Check out generationship.ai.

Christine: Yeah, there you go. And there's a lot of experimentation that's happening in terms of like people building.

I mean, a lot of people are building sort of like vertical agents, like agents for doing recruiting, that kind of thing.

And in general, like I guess, the lens that we have onto a lot of what's happening in this AI space is because we're providing this data that like all of these companies need.

We do have a couple sort of AI features that we provide, in that, like, we have, for example, functionality in our APIs to extract structured data out of email communications, and then we have an API that allows you to capture video recordings and do the transcriptions.

And obviously, the transcription part is AI-powered, and the summaries, and things like that.

That's something that LLMs are very good at and kind of the current tech does a good job at.

But the place where our customers are innovating is like applying sort of like these basics and taking that contextual data into a specific vertical, specific workflows, and kind of like going through the same playbook that everybody went through with vertical SaaS where I would say that like this is a trend that started maybe five to 10 years ago where people started building SaaS applications to encapsulate the best-practice workflows for different types of businesses.

And today, you have like a vertical SaaS app for everything, like running your accounting business, running a pizza shop, running a yoga studio, like you can just buy an off-the-shelf app that does all the right things for you to run that kind of business.

And now, the experimentation place that people are having is like, "Oh, what if we do that, but like super sort of AI-powered, and with like a different user interface because we're going to be like offloading sort of more tasks or we're going to have a different user interface than kind of like a pointy-clicky, web UI-type thing."

For us, because we sit in this kind of API space, it's mostly not our job to do pushing.

Our job is to understand like: What are the things that are getting uptake, and where is the demand, where's the market going? And then provide people the building blocks that help people go there.

So, there's kind of a delicate balance of making sure that we have our timing right for helping people keep pace with the things that are becoming user expectations.

But it's hard for us to kind of go ahead of the ball and say like, "You should be building this type of thing," because every new feature starts out as like the special sauce for some location out there.

And like we don't want to be the special sauce, we want to help you focus on your special sauce.

Rachel: A feature request, if I may.

Christine: Okay.

Rachel: I would love you to transcribe all of my conversations, extract action items for me, carve out some time on my calendar, and then hire an intern to complete those tasks and email me when it's done. Could that be arranged?

Christine: You could totally build that with Nylas. And I think a lot of people are trying to build different flavors of that. We'll see which one wins.

Rachel: We are continually told that AI can help us reduce the busy work. But in practice, is there resistance to adoption? Are you seeing people use all of your features, except the ones that are AI-enhanced?

Christine: Definitely sometimes.

I mean, we get, for example, security inquiries about like: Are you using any of this data to power AI? Is there an on/off switch for like AI stuff? Can we opt out of AI?

And obviously, like I mentioned before, like we don't give you AI if you don't want AI. Like, this is infrastructure.

But I think that there is hesitation around like data loss and a lot of sort of security stuff, and there are also like some, just like very accuracy-sensitive workflows that like there shouldn't be like a 99.9% accurate robot or an 80% accurate robot as a part of kind of how that stuff is going.

So I think there's a big place for human in the loop, and I think the hype is definitely a bit beyond the actual usefulness in a lot of places right now.

Like, if you read the platform formerly known as Twitter and sort of like the VC world, like you'd think that you could just offload all of your work to an agent, and in two years, nobody's going to have a job anymore because it's all going to be automated.

But if you go and talk to anybody in the real world, there's so many details to get right, and I think the reality is somewhere in the middle.

I do think AI tech is something that is going to really change the world in a major way over the next five to 10 years. But it's probably going to be a little slower than all of the hype people think at the end of the day.

But, I don't think that like in five years from now, like we're all going to be interacting with computers primarily through pointy-clicky interfaces, and I think that's actually a good thing.

I don't want to spend all this time like with like these sort of very unhuman-like interfaces for interacting with technology.

I think it's amazing that voice interfaces have gotten this great leap forward.

And like if I have a question, I'm like in my kitchen, I can like yell at my little robot in my kitchen and say like, "How many ounces are in a cup?" and get like an answer right back right then.

And I think that like technology is going to continue to adapt to be more human-like in a way that makes us feel less like technology is this specialized thing that I have to go into my technology brain worlds to use. It should just be something that organically is a part of our lives.

Rachel: Yeah, it should be supporting us, it shouldn't be our master. And that's the dream, isn't it?

Like we spend a couple of hours a day doing really interesting intellectual work, and then we knock off, and go and cook a really nice meal, and sit in the garden with a library book.

I mean, isn't that what we're building all of the automation for so we don't have to work all the time?

Christine: No, it's just a little depressing to see how, it seems like, some of this tech is eroding artistic positions faster than like doing my laundry.

Rachel: Zooming out a bit, what excites you most about where we're heading with AI and enterprise software, and what still worries you?

Christine: Yeah, the thing that I'm most stoked about around my working life is that I think AI is going to produce smaller, more empowered teams.

Like the kind of mid-20th century model of industrial capitalism, where we like design processes where every person is responsible for some tiny chunk of the work. It's not that fun, it's not empowering, it's not satisfying to us as human beings when we show up and we don't get to see kind of the end results and the impact of our work.

I do think that is an amazing thing that we are moving towards a world where small teams can get the same amount done as a much larger team in a former era, and that's hopefully going to make just people's jobs more interesting and engaging over time.

I've definitely seen that, even sort of to a smaller scale today, that like teams today can get more done than a team 10 years ago could.

And I think it's really great because I want to spend time building things and making things and seeing impact.

And personally, I'm not psyched and like super interested in the politics that happen when you sort of have to get things done with a large amount of people.

And we just weren't designed to operate in huge groups of people, where we like don't really know all of the people in that group.

Rachel: Keep it below Dunbar's number.

Christine: Yeah.

Rachel: The size with Neolithic village.

Christine: Makes things a lot easier. So I think that is something that is really exciting to me.

The thing that I'm most worried about is: What happens when we can't tell what information is real or not?

Rachel: Mm-hmm.

Christine: And actually, I already see that in myself where like I like don't trust videos, or text, or like stories that I see online because I'm like, "Oh, this is trying to emotionally manipulate me, and like this is happening someplace really far away, and like I don't know if that's real or not, or like what the perspective is of whoever created this content."

And I don't want to let that impact me because it doesn't seem worth it.

Rachel: It is surprising to me as an arts grad that media literacy has suddenly become this very obviously quite rare and valuable attribute because people present me with like writing that was done by AIs and say, "It's really good, isn't it? You couldn't even tell it wasn't written by a human."

And I'm like, "On what world can you not see how bad this is?"

Christine: Yeah, uh-huh.

Rachel: There was a paper that had people assessing real versus generated poetry, and they were like, "Oh, I like the bad poetry better," and I'm like, "Back to the library with you for 10 years and you're not allowed to leave until you understand what poetry is."

Christine: Yeah, I'm mostly thinking about things for like current events and political stuff, like-

Rachel: Yeah, yeah, the disinformation. But it is the same skillset, like being able to know if that is something that a person would really say, or if that is a story that sounds plausible.

Christine: Got to put on your propaganda filter.

Rachel: Yeah, it's all going to be fine somehow in some way that we can't see yet.

Christine: I mean, I'm generally on team optimism.

I think we'll figure it out, but that's the thing I'm most annoyed about is like I just don't care about like things that happened so far away that I don't have any personal connection to them anymore because I don't even know if that story or that thing was actually a real thing that happened.

Rachel: Yeah, and that's a real loss, that sense of connection that we had when, in the early days of the internet, where we could actually form personal relationships with people we would never otherwise have met.

Christine: Mm-hmm, mm-hmm.

Rachel: It feels like that window is closed.

Christine: Yeah, it almost makes it seem like we're moving more towards a world that is sort of more locally-based again because without the computer to sort of intermediate, like things just feel more real.

Rachel: Yeah, I do know my neighbors better than I have in a long time so that's... It's been another interesting thing.

Spang, what are some of your favorite sources for learning about AI?

Christine: Honestly, one of the places that I still find the most news is the platform formally known as Twitter.

Rachel: Mm-hmm.

Christine: I think you have to curate your feed a lot. But it still seems like the place where like people who are innovating post things the most.

Rachel: Are you on Bluesky at all?

Christine: I am, but I don't know if I need to like follow lists or something because I'm not getting the news in the same way there.

Rachel: Yeah, I think it's very good in some areas. I think it's been...

There's a lot of journalists there and a lot of writers, and certainly when there's an earthquake in San Francisco, you see it the way you used to see it on Twitter.

Christine: Yeah.

Rachel: And I'm seeing a lot of stuff that would've been on Twitter in the past now on LinkedIn, like a lot of much more technical, and-

Christine: I have seen that as well. It seems like things have fractured a bit, and it is annoying that I do feel like I have to be in multiple places at the same time.

Rachel: Only so many hours in the day.

All right, I've bought your platform, I'm going to make you emperor of the galaxy for the next five years.

Everything's going to go the way you say it's going to go. What does the future look like?

Christine: I'm ready for solar punk towers with tons of plants and trees, like really big trees everywhere that are just like majestic and like sort of intertwined with the buildings, and people are focused on connecting and doing art, and just being.

Rachel: Ecotopia, absolutely here for it.

Christine: I love plants.

Rachel: I love plants. If you had a generationship of your own, a starship that can go on journeys to take more than one human life, what would you name it?

Christine: It's going to be called Yggdrasil, which is the World Tree from Norse Mythology.

I have Norwegian heritage, and I've actually recently been reading a book about some... The history of the Viking Age, and I really just love all the mythology there.

And there's this myth of the tree that sort of connects all the worlds, and the underworld, and the current world.

Could have been come up with by looking at the Milky Way.

Rachel: Hmm, that's beautiful.

Christine: In the dark skies. And I just think that's really lovely.

Rachel: In the forest just north of where I grew up in Sydney, there is emu carved in the rock by the native people, and it was realized a few years ago that, at the right time of year, it lines up perfectly with a dark cloud in the Milky Way as seen from the Southern Hemisphere in the shape of an emu.

So, the Milky Way is a big part of my mythology as well.

Thank you so much for coming on the show. Always a delight to chat. And good luck with everything.

Christine: Thank you. Great to be here.