Devin’s 80% Moment: Background Agents, 7x PRs, & End of Hand-Held Coding — Walden Yan & Cole Murray

When people think about the ability of an AI to run your app and test it, I think they actually over-index on the computer use part of it because computer use in my mind is the literal, "Okay, you want... You know a button you wanna click. Can you emit the right coordinates to go click that button?" I think testing is actually a really interesting-

Usage

... problem-solving-

Yeah

... uh, challenge for these AIs because if you wanna do arbitrary testing, like imagine you make a change that spans the front end and the back end. To actually test that change, we have to reason through what-- how do you first run these applications to orchestrate with each other with the right version of the code? Then, okay, how do I trigger the feature or how do I make the thing actually happen?

That is where we spend most of our time.

Before we get into today's episode, I just have a small message for listeners. Thank you. We would not be able to bring you the AI engineering, science, and entertainment content that you so clearly want if you didn't choose to also click in and tune into our content. We've been approached by sponsors on an almost daily basis, but fortunately, enough of you actually subscribe to us to keep all this sustainable without ads, and we wanna keep it that way.

But I just have one favor to ask all of you. The single most powerful, completely free thing you can do is to click that subscribe button. It's the only thing I'll ever ask of you, and it means absolutely everything to me and my team that works so hard to bring The Latent Space to you each and every week.

If you do it, I promise you we'll never stop working to make this show even better. Now let's get into it.

All right, we're in the studio with Walden Yan, co-founder of Cognition, CPO.

Yeah. Happy to be here.

Which is a cool title. Um, yes, and one coiner of context engineering.

Yes. Yes. Although I, I think there are many people who used the terms in various ways beforehand, but, um, I, I, I did find that people both internally and externally enjoyed the upgrade from prompt engineering or-

Yeah

... you know, mo- model wrapping into maybe a more thoughtful way to build agents.

Yeah. For, uh, for those who haven't caught up on that, uh, I have on screen the don't build multi-agents post, which you should go read, read on and we might refer to. And Cole Murray, who created OpenInspect.

Great to be here.

Okay, so let's talk about it. Everyone is building their own Devins. Um, what's going on?

Yeah, so I think the engineering world is kind of waking up to this idea of background agents, cloud agents, uh, whatever you'd like to call it, and I think we saw a shift around the December timeframe of 2025 where the models Opus 4.5 and GPT 5.2, they reached a capability where we moved away from kind of handholding the model and being able to actually more or less autonomously drive the model.

And what I mean by that is that we could pretty much go from a specification to a completed pull request, assuming the spec was good enough, uh, with very little friction. And that paradigm alone I think changed a lot of how we interact with agents, um, and kind of opened this world where background agents became more practical.

I think for, for Cole, everyone experienced this in December, but I feel like there was just this y- increasing ramp, right? Like, um, there was this... the moment which was I think Sonnet 3.7 where, like-

Yeah

... you guys rewrote Devin in one night or-

Yes. Yeah, yeah.

So describe 2025 or, you know, how, how it felt from your side.

In retrospect, you know, we always thought it was ramping up, but then even now over the last three, four months from today, we've... this has been ramping up even faster. So it's, it's almost funny to be talking about how like big of a leap Sonnet 3.7 was, and we... Honestly, a lot of it was stripping out parts of Devin that were no longer needed with that jump in in- intelligence.

But I also just think that a, a lot of the recent leaps, uh, especially, you know, you look at like models like Opus and the latest GPT models, they are reaching levels of autonomy where people are actually finding that they actually can just be hands-off. And people who were once debating, "Oh, you know, do I need to be in the weeds with my model in the IDE?

Um, can I just completely move it off into the cloud?" That's a, that's a more serious conversation, and we've seen that in, in all of our growth charts. Um, internally there's this funny graph where our usage has, uh, of PRs or our merged PRs has grown 7X since-

Yes

... I, I forget what it was called.

I think Dev, uh, maybe tweeted that.

Yeah.

Uh, yes.

Um, it grew like-

I'm looking

... 7X over like the last I think it was like two months, three months, something like that. Uh, and then you see our engineering headcount growth has like gone up by like 10% or something.

Like we were, we were afraid-

Yeah

... to release this.

This is-

So, so this is Devin commit percentages on all Devin repos, uh, was 16% in January and now 80% in March.

Yeah.

It's like, uh, it's a big shift right now. Um, and so it, it makes sense that a lot of people are now thinking about, you know, buying Devin but also maybe like, you know, trying to build their own and, and there's, uh, lots of... I, I have a lot of fun building Devin, so I can see why other people would want to build their own cloud agents as well.

Yeah.

Well, maybe it's, it's good to hear like what, what initially inspired you to try to build OpenInspect?

Yeah. OpenInspect came about, uh, through primarily my clients observing how they were using tools like CloudWeb, um, OpenAI's Codex at the time, and seeing some of the friction that they were having with it. Um, primarily the CloudWeb, uh, was being used through Slack, and a big issue they ran into was that the sessions that were launched were specific to whoever called it via Slack.

Mm.

And so if a PM was the one who invoked the session and they would then go to pass context engineering, engineering can't see the session. And that in itself was kind of a deal breaker because the PM, "Hey, engineering, can you jump in?" But there's nothing to jump in on unless they're copy-pasting out or, you know, the single response that came back.

Um- And so kind of seeing some of these problems, I had built a similar kind of architecture internally, um, just to experiment with, um, kind of test out different ideas as this trend of moving off of local hosts was starting to kind of become, um... And as Ramp released their blog post, I had a lot of the pieces for this already in place, um, and just thought it would be kind of funny to, uh, see what Claude could do just purely from the blog post.

And, uh, on my X account, there's actually kind of a, a thread of where I live tweeted, like going through this-

Oh, wow

... uh, comparing GPT and Claude as both of them are going through it.

I- like, on the announcement thing or something else?

Uh, right after it got released.

Okay.

Uh, we can put it in the show notes.

Okay. All right.

Um, yeah, it was helpful that I had already kind of knew how to verify the system. I knew what I was looking for. I think Ramp did a great job of really illustrating, uh, the technical aspects of how to build something. Um, it was much more than just kind of like, "Hey, we built a great system." It was, "And here's how you can build it, too."

And so, um, I resonated a lot with that, just with the problems that I was already seeing. And I thought that, uh, looking around, I didn't really see anything in the open source community that, um, kind of met this type of system. I think there's a lot that run, uh, in local hosts like Superset, Conductor, um, and many others, but nothing that was actually running in the cloud.

And so, um, I built it, and I thought it was interesting to just open source it and allow anyone to then have a foundation that they can mix and match on top of.

So literally after Devin was launched was, there was OpenDevin-

Yeah

... which became All, All Hands. Uh, I don't know if you tried that or...

Yeah, I was gonna say, one of the things that interested me a lot with OpenInspect was like you didn't try to go make it then something you, you monetize. Um, um, there are a lot of, I think, these open source projects would then go, and they try to, like, raise VCs-

That's why no OpenDevin. Yeah.

Um, yeah. And, uh, how did you think about that? I, I thought that was very interesting.

I thought, and kind of just what I had seen across my clients, was that having a background agent system is going to become a critical infrastructure within their company. And so because of that, I think that I wanted to open source it so that they could fork it and put in whatever customization they wanted. Um, to that question though I get asked all, like, "Oh, are you gonna raise?

Are you gonna turn this into a service?" Um-

I'm sure you've gotten offers.

Uh, but, uh, primarily I don't want to do that for a few reasons. One, I think that I don't wanna compete for like $20 a seat. I think that that is just a really difficult business. I think it's very easy to copy the main pieces of it. I mean, again, like I built this fairly quickly. And I think because you are not owning, I guess, the entire stack, it's hard to monetize.

You have money being made at the sandbox layer with Daytona, E2B, many other players. You have money being made at the model layer. Um, and you kind of sit in this weird in-between gray area where, what are you actually selling? You're selling, I guess, the infrastructure. You're selling, uh, the integrations maybe.

Um, let's ask the guy. What are you-

Yeah

... what are you selling?

Well, yeah, there's multiple layers to this in, in practice, and actually, it's funny you mentioned the infrastructure, 'cause when we got started building Devin as well, uh, we had to go figure out how to make the infrastructure as well because, um-

You had to build this b- two years before everyone else, you know?

Yeah. Yeah, exactly.

Yes.

And including like the modern side. Yeah.

It was not, it was not very polished at the start, like when we just built it off of raw VMs from cloud providers like EC2. The boot up time was so slow, I think, uh, and especially the, then, like turning off the machines, saving them, and then to be able to bring them back up again when the, when you want Devin to wake up again later.

Uh, it would just be out cold for like 10 minutes because that's just how long these systems took. They were not built for this repeated down and up usage. And so we actually had to go do all of that. Um, and as a result now, one thing we offer when we go and sell Devin to people is, you know, you don't have to worry about all the compute side of things.

We'll make it work. We'll make it work in your cloud if you want it to. Um, but as- aside from the product, and I, I wanna go into the agents and, and, and the tuning of the intelligence part later, but I think a big part of what we do at Cognition as well is to just make sure that your company learns and uses and adopts these coding agents.

'Cause I think for especially the largest enterprises in the world, you find that there is a lot of people who want to move over to using AI for their day-to-day workloads. Um, but because of the way projects are planned, because, um, not everyone is literate in using AI in these ways, uh, having a team of engineers who can actually go in and onboard you, set up all the integrations you need, the automations you need to really get to that level of, you know, leverage with AI is super helpful.

And so-

Total

... we do that. We, we show up as sh- thought partners to the customers that we work with as well.

So let's talk about like architectural stuff. I, I think that's always, uh-

Yeah

... uh, uh, that is something that was the topic of conversation between the two of you. Is this, uh, sort of like the mental model that you wanna start with or something else? I, I, you know, I'll just kinda leave the floor open to you guys.

Yeah, I think, uh, maybe we can start here as just kind of a, a general, what are the pieces of a background agent system-

Yeah

... and then maybe we can go into some of the nuances of, uh-

Yeah

... decisions that you can make.

But I guess, like, also, like, what, maybe what Walden is saying is, is the agent is kind of like in this open code box, I guess.

Mm.

Right? Like this is infra, and then there's, that, that's the agent. And you had this discussion about whether you put the agent in here or in-

Yes

... out, externally. Can you sort of tease that out?

Yeah.

In a background agent systems, you have a decision to make of where the agent is actually going to run. This is typically described as the harness-in-the-box or out-of-the-box.

Yeah.

With running the agent in the box, uh, you're making some trade-offs by doing that. The negative trade-off you're making is primarily security. Because the agent is running in that box, unless you otherwise design it, all of your secrets need to go into that box as well. And given the nature of AI, it can be unpredictable, and you could very easily end up accidentally exfilling your secrets- ...

um, or, you know, other kind of unintended behavior. Now, the out-of-the-box is the idea that we are going to have the actual agent running not directly in the sandbox, and we'll have, quote, unquote, "the brain of the agent" running in some type of worker, uh, control plane. That sandbox then is going to serve as the hands where the brain is basically operating and making tool calls into th- that environment to manipulate it.

I guess other trade-off that you're making between the two systems is that, um, in my opinion, running it out of the box is much more complex because, uh, you have state that has to be managed, whereas if you're running it in the box, uh, all of the state of that agent is actually in the box. And yes, it's...

you could persist it elsewhere, but it's all kind of localized, and you have less concerns to worry about.

I think a lot of that, what you mentioned, is why we actually, from the start, built Devin to re- what we call separate the brain from the machine. Uh, the other thing that this allows you to do is reuse any existing infrastructure you have for dev boxes-

Yeah

... perhaps.

Sure.

And so you don't have to worry as much about making a new type of dev box that has all the dependencies the brain needs, as you mentioned, the secrets the brain needs as well. Um, like, one thing that we've seen some customers run into is, like, you have a GitHub app, and you want Devin, your agent, whatever, be able to interact with GitHub through this application.

But then you have different users with different actual permissions.

Sure.

If they are all interacting through the same GitHub app and there's no actual, like, separation between the system that decides, uh, what it does and the actual secrets on the machine, then you kind of run into an issue where, okay, it's, it's hard to do the separation. But in practice with Devin, it's much easier because we just say, whatever you put on the machine, that is, like, kind of the scope of basically what the user is free to do, what the agent is free to do.

So only put the most scoped secrets on that machine, and then the brain is fully not, not accessible from the machine. So you don't have to worry about messing with the, any of the most secure parts of the brain if the user is free to do whatever they want with the machine.

I was gonna just bring... I had this, like, chart from OpenAI, uh, where I don't know if this is, like, in the box, out of the box. Uh, th- that is something that they do use to describe it. Uh, and then also recently, Anthropic did, like, manage agents-

Mm-hmm. Yeah

... which is, this is their thing. I don't know. It's all, it's all variations of the same pattern, right?

Yeah. So this would be out of the box.

Yeah.

Which, like, is, is preferable for them because it's less work?

Uh, I would say it's more work.

It's more work?

But it-- In my opinion, it is the better architecture of the two.

Okay.

It's just, uh, you're taking on a bit of complexity by doing that.

One thing I've not seen a lot of other players do well is how do you manage what's actually on the box? And this can be complex for many reasons. Like, let's say you have a big repository that's changing and updating a lot with changing dependencies. How do you make sure that the working environment of the agent actually stays up to date, has all the credentials it needs to, let's say, run the app and test it, and all the things you want your autonomous-

So it's repo setup.

Yeah, exactly. So in, in- ... internally at Cognition, uh, we call this repo setup. The hardest part of it's been a perennial problem since the start of the company of how do we help people get this set up? 'Cause not everyone just has, you know, uh, working cloud environments working out of the box. A- and d- do you find this to be a common problem with-

Yeah, how do you solve it?

... some of your clients?

Yeah, this is a very common problem, and, uh, through my consulting, this is a lot of what I help teams do. Um, a lot of teams don't really have great developer environment setups, if any. Um, a lot of the times it's, "Go talk to Bob and get the secrets," and that obviously doesn't work when the agent needs to actually set this up.

And so a lot of that, uh, most teams are using Docker Compose or some type of microservices. And so-

Even in prod?

Not in prod. Um, like, with the OpenInspect, you are using this primarily to interact, um, and make code changes. There is other use cases, but, um, you can hook, whether through CLI, MCPs, other tools, um, you can then hook that into your production systems primarily for, like, SRE type use cases. But you are not, uh, necessarily, like, trying to test your prod internal microservice through the system.

Yeah, and you mentioned Docker Compose. I think one direction we saw some of our friends take early on was, um, using Docker containers as the level of ex- abstraction for their models. Um, there's lots of reasons I think why Docker containers are not great. Um, one thing is, like, Docker container's not really a true security boundary, for one.

Uh, but the other is, like, if you are running real applications, a lot of times those applications use Docker, and then you have to think about Docker and Docker, which is, like, really weird.

Yes.

And so I think part of, like, the really hard challenge of getting VMs to work, why did we do that? Well, it was because we realized that you actually needed, like, full VMs to be able to do these types of things. And especially nowadays, where there's actually value in running the application and clicking around and sending you screen recordings of these things.

Like, the, the value just kind of, like, keeps adding on, on top of that. Um, but it, it is a decision I, I see people run into when they try to build their own systems is, oh, like, do we... Like, in addition to this, do we put the agent in the machine or out of the machine? Do we use Docker?

Do we use something else? What, what do you recommend people nowadays?

I think Docker is a good solution for maybe not running the agent, but running your infrastructure because that is more or less the same setup your engineers are probably already using. Um, if they're not, then I don't know what they're using. But, uh, they're probably already using Docker Compose.

Yeah. I, I've always had a small candle for web containers. I don't know if you guys have tried them before.

Oh, yeah.

To me, they were, like, supposed to be like Docker Lite.

It's like-

I don't know

... no, I haven't tried it.

Yeah.

Um, but yeah, I think any environment that you've set up that is a good experience for your developer naturally lends itself to being easy to set up for the agent. And once you figure out kind of that local developer story, um, you've kind of more or less solved the agent in a sandbox, uh, environment setup. OpenInspect does have hooks as well where you can, uh, run a setup SH script that will pre-install everything.

You can then pre-snapshot that build so it starts instantly, and then there is a, a second hook to actually then, like, restore the state of the sandbox when it comes back. And so you can already have all of those microservices running and basically get the same experience that you would on your machine within the sandbox.

Another thing that we've been thinking a lot about is kind of like different VM service offerings. Have you had customers where they needed, like, macOS specific VMs or like Windows specific VMs?

Oof. Not yet.

There are, like, many technologies in the world that only work on specific types of machines, right? If you're building a .NET application that has to run on Windows or like, you know, maybe more commonly if you want to build iOS or macOS-

Yeah

... does like that machine-

Does Cognition support-

... machine

... choices like that?

The fundamental architecture we do, because we do the separation, it does support, but the actual work in progress is happening right now on these. Uh, another thing that we've actually recently added support now for, it's in beta, is doing Android development. To do that, we needed to support, I think, nested virtualization within our machines because the, the VM itself is like a, is a virtualized Firecracker instance, and then you have to then run another Android emulator inside.

Um, and you know, there's like weird performance issues that like, you know... It, which is why it's like still in beta. We have to think through these problems, but it unlocks a lot for, for anyone who wants to do Android development.

I was trying to find like a reference video for the, the testing thing. I couldn't find it.

Yeah.

But, uh, uh, I, I think you worked on the testing, uh, capability. Why call it testing and not like computer use or... I don't know. It's what's the general-

Yeah

... category of problem?

I think that when people think about the ability of an AI to run your app and test it, I think they actually over-index on the computer use part of it because computer use in my mind is the literal, okay, you want... You know what button you want to click. Can you emit the right coordinates to go click that button?

I think testing is actually a really interesting, like-

Use case

... problem-solving-

Yeah

... uh, challenge for these AIs because if you wanted to do arbitrary testing, like imagine you make a change that spans the front end and the back end, um, maybe, you know, even some other like even d- more deeply nested service. To actually test that change, we have to reason through what... How do you first run these applications to orchestra with each other with the right version of the code?

Then, okay, how do I trigger the feature or how do I make the thing actually happen? Um, and this can kind of get arbitrarily hard, like maybe you have to be an admin. Maybe a certain thing has to be feature flagged on. Maybe, uh, you have to like run two sessions and then send us a very specific word into one of them to trigger specific behavior.

And figuring out how do you do that requires a lot of code-based context, requires, uh, a lot of orchestration that we've specifically done. And, uh, in some cases we found that you actually no one frontier model can actually do this full end-to-end task itself.

Ooh.

We've seen cases where we actually had had to orchestrate different frontier models together to kind of solve this problem together. That is where we spend most of our time when we think about this testing problem, not so much the computer use part. Computer use for what it's worth has gotten a lot better with, with recent models and it's, it's made that part of the job certainly easier.

Yeah, especially with like even 47, um, that they released yesterday, uh, apparently like way better in terms of the division stuff-

Yeah

... which is gonna be encompassing computer use.

Having evals for all these as well is something that, like, takes a while to build up, um, and having the evals be right is tricky as well. Do, do you ever see cus- like, you know, clients who are building their own agents have to start standing up evals to make sure things don't regress?

Not so much evals in the traditional sense, but, um, specific to the testing part, that has just gone in. Um-

Yeah

... I just added support for screenshots.

Okay.

And, um, in theory, you can also do video. I need to put in a plugin to do that, but, uh, they do kind of show up natively, and it was a very heavily requested feature, um, especially after Cursor's recording came out. Um, I think that was very enlightening for everyone of like, "Oh, this is a very good feature to actually have."

Um, I think with Devin you guys have had this for a while.

Yeah.

You know?

She was first.

Oh, yeah. See how screenshots work.

Yep.

Um, yeah. I, I don't know if there's any- anything, uh, super not obvious. It, it's kind of like once you know what pro- feature to build, you can just kind of prompt it and it'll-

Yes.

Most work.

Yeah.

Yeah.

Um, I think to Walden's point though, the computer use is kind of a subset of the larger testing problem, and I think that that's very specific to the code base that you're working and it's not something that, uh, you know, out of the box that you could just solve it. The... You do need the code base context to actually know how to test it, and I think in the case of a background agent system, you fortunately do have that code base locally that you know what is changing and could then inspect it and use that to drive the model.

Yeah, yeah. Uh, for those who haven't seen it before, this is an example of how it works. Like, you, uh, after the PR is done, you click testing approved, and then it sends you back a video. Uh, what I really like is that it labels, um... It's very small here, but it actually labels what it's testing, um- And then it, and then you actually see the, the cursor and everything.

Mm. Yeah.

So I don't know, like, uh, yeah, the engineering in this, like, just whatever, whatever you wanna show. Like, 'cause this is, like, you know, this is one of those, like, oh, feel the AGI moments, right? Like-

Yeah

... 'cause once I look at this, I actually don't e- I ca- I, I wish I can just merge inside of Slack instead of going to GitHub 'cause I don't need to see the code. I know it works.

May- maybe a new feature in GitHub.

Yeah.

Yeah. Um, the annotations at the bottom was also a big difference for me when I, when I added those.

Yeah, 'cause it's like what, what am I looking at? What are you trying to demonstrate?

Yeah. Exactly. Um, there's a surprisingly long tail of small details that ends up making a big difference for this kind of end metric of, like, how fast do you actually merge the code in. One experience that we spent a lot of time tuning early on was what is the right experience on GitHub for these tools.

Sure.

Because I think, um, most tools out there, when you build the agent, you'll think about, oh, like it'll create the PR for you. We, we try to take that a step further and say, "Oh," like, "what if we actually made sure you could interact Devin, with dre- Devin directly on GitHub?" And so we made sure that you can comment on GitHub, and Devin would actually receive those comments and address them back.

But there's actually quite a bit of tuning you have to do here because you can imagine that actually, like, we recently have Devin Review, for example. Um, Devin Review will post comments on his own PR-

Yeah

... and then Devin has to then go-

He answers his own comments, which is really-

Yeah

... really loopy. Uh, so like yeah, I, I like that it just updates here that it, that I have commented.

Mm.

But, uh, usually it's just me saying like, "Hey, merged. Uh, fix any merge conflicts."

Yeah. The... So when Devin fixes his own comments, you might be scared that, oh, like maybe I'll infinite loop.

Yeah.

But we put a lot of work into making sure it doesn't, um, both by making sure that the comments are high signal, but also that the agent is thoughtful about what comments it immediately goes and tries to fix and what comments it's like, "Wait a second, I think you're wrong." Actually, that's one of my favorite moments is when Devin tells me that I'm wrong when I-

Nice

... I try to get it to do something different.

Yeah.

But tuning that behavior, like, actually makes a big difference in terms of how useful the actual GitHub experience is.

Yeah.

Yeah.

I think to touch on that as well, I think having the AI reviewer integrated into the system is a critical part of this background system. Um, OpenInspect does have that. It has a GitHub code reviewer that you can control the prompt. Um, it does do comments as well. It doesn't do them automatically yet. Um, the capability is there, but it's not fully used.

So you have to ask for it.

Uh, you do, yeah. You can tag it on GitHub, and then whatever you named your, uh, GitHub bot, it will then follow up on it.

Okay.

It will then, uh, if you have merge conflicts or whatever you have asked it to resolve, it will then resolve it, but it doesn't do it automatically yet.

Well, I'm curious, what is, like, the most common thing that people end up requesting, uh, that they still need on top of OpenInspect when you help them go implement it?

I think a lot of it comes down to actually integrating it into the company. It's one thing to kind of have the background agent system set up, but if it isn't actually integrated into your larger ecosystem, it isn't that useful. I mean, it is useful to be able to kick off sessions, but what we really want to be able to do is hook it into all of our other systems, whether that is the production database with read-only credentials, the logs, um, a Confluence or internal knowledge base system.

I think that is where I see the huge leap for companies. And, um, that can be a challenge for companies as well who are maybe not familiar with exactly how to approach it, um, especially if they're in environments that have more compliance type things where, um, access control can be pretty big and how do you deliberately think about these problems, um, I find to be, uh, kind of one of the problems that comes with a system like this.

Yeah. The, the thing we found is... So, like, MCP is, um, obviously it has been like this, like, really big explosion of, oh, you can go, like, integrate it with all these different things. Uh, but to actually get the integration right and the r- get the right experience, oftentimes we found that we had to go build our own ad hoc things.

Um, I think Slack is a great example of this.

Mm.

Like, you could give your agent a Slack MCP and okay, it can post messages back to you on Slack. Um, but we actually use Devin like a coworker in Slack, and that's how it's been built from the ground up. But to do that you actually need to, like, support webhooks that come back, right? And then Devin has to respond in a natural way, and then hopefully don't spam your, your threads too much and annoy the people in your company.

So you gotta tune that experience just right. Um, especially when there's a lot of back and forth, like we find that we actually have to go beyond the simple MCP integrations in these places.

I just pulled up the MCP marketplace. I know this is a fair amount of work. Um, I, it... I mean, is, is the answer to eventually take first party control of all the top MCPs? Like, is that the-

Uh, I would love a world where you could have something that's more expressive than MCP, okay, that kind of like goes both ways. Like, not just a set of tools, but, like, a proper system that interacts back and lets it-

Yeah

... have the right experience with all these interfaces.

So there actually is sampling in the MCP spec-

Yeah

... but nobody uses it, right?

And so I think that's the other part is, like, actually we found that when the MCP spec starts to get too complicated, um, it starts to lose its original promise of-

Yeah

... being like a simple one-step connect. Like now w- then we have to go figure out how to support all these different variations of things and-

Okay

... it starts to look a lot like just building their first party integrations in a lot of these cases now.

Yeah, I think it matters, too, how critical it is to your company, right? If this is something that nearly every session is going through, it probably makes sense to own it so that you can make optimizations on top of it-

Yeah

... versus just whatever is off the shelf.

Yeah. Awesome. Other than MCPs, what else... Uh, sorry, well, I don't know if that's narrowing in too much on, uh, integrations. Uh, but what else? Like, what, what other elements of building OpenInspect or Devin that you guys- Really sync on

Yeah. I think, uh, a problem that comes up very frequently is this idea of memories or knowledge base. Um-

Oh, boy. Yes. How do you solve it?

Uh, so not solved yet, uh, is the short answer.

Uh-huh.

Um, it's something... There's a open issue for it, someone asking about it. Um-

Okay. There, there's... I, I... Uh, D Wiki hasn't indexed anything about memory yet.

Um, how I'm seeing it solved across my clients is primarily through skills. Um, I find that skills can be a good gap within that or updating-

Yeah

... Claude MD, but I think memory as a whole is a pretty unsolved problem, and it is why I've been kind of hesitant to add it. I think there is parts of memory and, um, that can be addressed, but I think as a whole it's a very difficult retrieval problem.

Oh my God.

Yeah.

RAMP didn't write anything about memory? I see zero search results.

Memory can be quite tricky to get right because it's the retrieval, but also the generation of the memories that can be really tricky. Like, you don't want it to just like-

Yeah. Okay

... remember very specific detail.

Wa- walk us through the Devin memory journey because-

Um-

... I know there's been a journey

... uh, the first version of memory that kind of, like, stuck around for a while was-

Yeah

... a system we have called Knowledge.

Yeah.

And the idea was we wanted it to pick up things over time and not need the user to be proactive about teaching Devin things. So, okay, any time you remind Devin, "Wait, no, that's not quite the way you're supposed to use Git," like, uh, we actually want Devin to say, "Hey, do you want me to actually just remember this for the future?"

And for you to just basically quickly approve or reject and for it to build up over time. 'Cause I, I find that, like, 95%, I think, or some crazy stat like that of the, of the memories that Devin has are all through these auto-generated things. Like, very few people actually just want to sit down and write big docs on-

Sure

... okay, here's how you're supposed to work with the, the technology, et cetera. The generation and the retrieval has been something that we've been trying to tune a lot over the years. Generationally, you don't want it to remember something like, like, if you asked one time to like, "Oh, please open as a draft PR," you don't want to be like, oh, everyone forever now it should get, get their PRs as draft PRs.

But you wa- you do want some kind of, like, come bear maybe you want to say, like, oh, Cole generally likes, uh, things to be created-

Yeah

... as draft PRs. Same with retrieval, like, you know, if you have thousands of these memories, how do you actually make sure they're retrieved at the right time? Um, and that can be quite tricky to do right without exploding the context with a bunch of useful... yeah, useless information. Surprising amount of just, like, eval work to just make sure that, like, memories, uh, remains a reliable system as new models come and go.

Yeah.

Do you have anything that you could share on, like, memory pruning? Um, and-

Yeah. Like-

... like kind of the temporal aspect-

Like deleting and forgetting

... of memory. Yeah, exactly.

Yeah, yeah, yeah.

Yeah. The, th- today, the, um... So the things they could do is it could edit memories.

I see.

And so if your memory used to say like, "Oh, Cole likes to open everything as like a draft PR," then you can imagine, "No, don't do that." And then it'll say, "Oh, do you want me to update the memory to be Cole now want everything as, you know, open PRs?" I think that at the same time, we don't know if this is gonna be the final version of the system.

Um, whatever we have here will probably, like, translate into the new system that we'll, we'll be coming out with. But I think one big difference between two years ago and today is these, these agents are really good at using anything that resembles a file system natively. And so part of us are, is thinking, oh, should we rebuild memories to feel more like a file system that we let the agent navigate on its own?

That, that's been an interesting exploration. Um, also similar ideas in, in the scale space.

I'm pulling up OpenClaw's memory thing right now. So memory... OpenClaw has, like, this, like, daily memory journal thing, right? And you can... I mean, that is a file system you can kind of grep through and is a source of truth. I don't know if it's the best. It's probably super noisy. But at least it... if you lose something, you can discover it or you can apply some kind of, um, forgetting algorithm to, like, more ancient memories that don't get recalled again or something.

I don't know.

One thing we've been trying to do to push the boundaries of how you use agents at your company is letting an agent basically have a very similar file, like a memory.md or something, and just kind of, like, be your permanent PM for a specific set of issues maybe. So we have, like, some-

Mm

... Slack channels internally, maybe a Slack channel dedicated to, uh, a specific product like Deep Wiki maybe. And you can imagine that, or you want a Devin that never stops, it's just always awake, but it has this, like, memory dock that it can just re-maintain for itself about, okay, what are, like, the number one priorities of what we have to fix and prioritize?

Um, who is responsible for some upcoming work? Maybe they'll even ta- Devin will even tag you on some recurring basis.

Yeah.

Um, and so it's been an interesting move to see, okay, how can we actually use Devin for more than just engineering? Can we actually upstream above the engineering process, and maybe it's just Devin creating tickets, which then maybe some humans do, but then maybe other Devins do.

Yeah. One of my more fun automations is go research competitors and just suggest stuff to me on a weekly basis. Uh, that, that's, that's the automation. I can't find it right now, but basically it just, like, look at competitors and suggest things.

Sure.

And here are three things that you've suggested that I don't want any more of, and you just kind of stick that in a prompt. But like-

Yeah

... I wish actually... So for, like, when I... f-for example, when I reject a PR, I, I wish that it updated memory so that I can then just not have to go up, go back and update the scheduled, uh, sync or like, you know-

Yeah

... feature, feature request.

Um, you know what? We might change it soon. I, I guess OpenInspect, uh, in the time you've been around, has there been anything you tried to implement, but then you had to, like, undo and, like, do a different way? Nothing yet, but something that is on my mind. Uh, the initial way that I built it was that each of the integrations kind of lives as its own package.

And so you-

Right

... have the Slack bot, which is what's handling the webhooks, and then is basically interacting with the control plane.

Yeah.

As I'm seeing the system starting to be more integrated, specifically with the GitHub bot integration, I'm considering bringing that all into the central control plane because especially now I want to start, um... And a request that I'm getting is the ability to monitor, um, the actual, like- Pull requests being merged, as well as just kind of tracking of-

Like, what do I have open?

Yeah.

Yeah

What do I have open? How many of these are getting merged? How many comments are showing up?

Yeah, yeah.

To just kind of understand the health of the system. And so in the case of a GitHub app, you only have one webhook. And so then it's a question of, do I put that webhook in that GitHub bot package? That's kind of weird. It doesn't really make sense to live there because that package is more for, like, the code reviewer.

Or do I, like, centralize it? Um, so that's something that's on my mind of, uh, making that decision. I think the other one we touched on earlier is kind of the harness in the box versus out of the box. Um, I think long term the architecture will eventually come back out of the box.

Yeah.

Um, some of the newer tools that I've added are calling back into the control plane so that you don't have the secrets in the sandbox. And so I think long term I probably will pull the actual, um, agent out of the box, but, uh, I think for now it's fine.

Just, um, a quick question on pulling the agent out of the box. Um, I, I'm... One thing I'm very bullish on this year is co- agents calling other agents or spawning sub-agents or-

Mm-hmm

... whatever you wanna call it.

Yeah.

Does that make it harder or easier? I can't tell. Because if the harness is in the box, you can just spin out more boxes.

Yes.

If the harness is outside the box, then you're, it's less easy because you are, you have a, you know, a unicorn pet of, of a, of a harness that's, like, living outside the box.

I mean, in theory it would be the same way, right? Whether, uh, one agent has launched many, uh, sub-sessions within it, um, OpenInspect, for example, can launch sub-sessions and actually create other environments and then monitor them. In the case where it is out of the box, uh, that would basically just be an additional session that's running, and so that session is also running outside of the box.

It's running in your worker plane, wherever you're running this. Um, and then you really just have to think about how does your top level agent then interact with it. I do think it can be more complex just 'cause again, you have now a more difficult architecture, but I think if you figured it out once, uh, it's probably fine.

Yeah. Walden, I- I'm just, uh, throwing it open to you f- in terms of like, uh, uh, I, I call this kinda like meta Devin management.

Yeah.

Which is like the m- Devin's calling Devins or Devin scheduling Devins or, uh, querying trajectories or anything like that. Um, what have you built or un- unshipped, anything?

I think one of the surprising things we've seen is that a lot of the ways that, like, these, like, separate agents work with each other, and you want them to, like, parallelize their work, has still mostly followed the same manager sub-agents regime. And a lot of people I think are excited about this world where you have swarms of agents that kind of, you know, talk with each other all over the place.

We've actually given Devin an MCP so that he can just go arbitrarily message other Devins- ... and create new Devins, et cetera. But I guess, like, it somehow c- creates, like, a, a really chaotic world in that sense. And so w- we've still found that most practical use on a day-to-day basis has been one Devin.

Single Devin. Yeah.

Figure out how to segregate the work and get, have other Devins work on it in, like, a relatively isolated sense, each with their own boxes-

Yeah

... not sharing machines, so there's, like, a li- very low room for conflict is kind of the, the regime that you have to create today.

I, I'll call out, um, the experiments from Cursor, right? Uh, this is Wilson Lin's work on single agent to multi-agent, and you're obviously famously on the side of don't- ... don't build multi-agent. But they went through the whole thing, uh, only to arrive at, uh, this

which is exactly what Devin has, I think.

I think there will be a revision to that post at some point-

Okay

... about-

Tell us about it

... um, I think multi-agents were very much not at all possible a year ago. You do see more multi-agent experiments today, but you can kind of argue, are they really multi-agents or are they just kind of just, like, tool calls, you know? Like, there are people who, um, will create sub-agents to go look for XYZ file, XYZ implementation.

Has really nice context management benefits because all of the tool calls and tokens that it spends then get collapsed back to just the answer for the main agent. There's a lot of benefits to doing this. We basically have Devin do this with DeepBooky, make a call out to DeepBooky, give you back the results, but that feels like a tool call, you know?

It's not like these, like, two collaborators actually talking back with each, back and forth with each other. But I think the thing that gives me the most bullishness that multi-agents might actually be possible is actually what I said earlier about Devin will actually sometimes tell me I'm wrong and push back, and I think that demonstrates a level of maturity and communication today that makes a multi-agent world possible.

Like, one, can two agents who have seen different information come back to each other and actually figure out who is right, what, what is the correct implementation? They're not just, you know, yes men. Claude, I guess, is like, uh, used to just say, like, you know, what is it? You're, you're right, or, uh-

You're absolutely right

... you're absolutely right. Yeah. Yeah. Um-

The... Have you seen, have you seen the Codex app-

It's over

... troll entopic? This is the Codex app. Um-

Yeah

... inside of Settings, th- there's a little, uh, there's a little Easter egg, right? So if you go to, like, their, their themes or appearance, right, there's all these, like, color codes, and the top is absolutely, and it's Entopic's colors.

Which is such a troll. Anyway.

I love that Easter egg. Did you discover that yourself?

No. It was, like, someone, someone was, uh, tweeting about it- ... and I was like, I was like, "Is this true?" Because, like, sometimes people just tweet stuff to, uh, get a rise out of you, but, um, yeah, there you go. Uh, Entopic colors.

Yeah. Um, so yeah, we're, we're, we're out of this regime where, you know, it just says, "You're absolutely right," and they can have real conversations and real back and forths. Um-

Yeah, you can prompt it as well to be more adversarial or whatever. Yeah. Okay. Yeah. That, I mean, to me, that is more intelligence, right? Like, that, that is, is not just something that's, like, a dumb tool. It's actually pushing back on you.

Yeah.

Like, yeah.

Um, when... Y- you mentioned, of course, the blog posts.

Yeah.

There was one blog they had where they s- had a swarm of agents together and built a browser-

Yeah, yeah. That, that was... I think that was the one

You can have like a-

I, I think it's the same one

... a version of it.

Yeah, yeah.

We found a s- surprising success of, like, don't do a swarm or anything, just have one Devin, you know, it does its own context of mention. Just let it keep running for a while and give it some crazy tasks. I think we asked it to, like, rebuild, like, a Windows OS system.

Yes.

And it managed to do it just like, you know, go- going on for long enough. It's just-

Was this Andrew's thing?

Yeah, yeah.

Okay.

Um, there were lots of demos that we ended up not posting 'cause at some point we'd just be posting way too much, a bunch of like- ... kinda demos. But I love that because it kind of shows that I think the multi-agent thing still has, like, a bit of exciting sexiness to it, which is maybe still beyond still, like, the actual delta it adds to the capabilities of these systems.

But it's absolutely the future. I think we're, you know, we- we're heading that direction and, and we can see the progress being made there already.

Yeah. If I were to, uh, make one super minor pushback because I don't feel that confident about it yet-

Go for it

... but I've, I've had Ryan Lapopolo from OpenAI on, on the-

Yeah

... pod and he's a super slop cannon, right? Oh my God, that's my coding agent being done. Uh, I, I, yeah, I downloaded this like, uh, Peon Ping. I don't know if you guys have heard this. It, it takes like, uh, sound packs from popular games like-

Mm-hmm

... uh, Command and Conquer and, like, Warcraft, and then it plays it whenever it's done.

Nice.

And so it's like, "Work, work," or whatever, like, "At your command," or something. Anyway, what I got from, from the Cursor code base and from, from Ryan's thing was that there's a slop cannon approach where you try to loosen the single agent's-

Yeah

... uh, bottleneck, and I feel like that is, like, probably an, a very important thing to try to figure out. I don't think anyone's, like, really solved it-

Mm

... because then you just have more reviewer slop on top of the agent slop-

Yeah

... to, to try to wrangle it all. Uh, Ryan will probably very strongly object that I, I say that he hasn't solved it, but he thinks he's- ... he thinks he's completely solved it. But, like, I think it's still im- I think it's, like, very important 'cause, like, that is a bottleneck, right? I, I feel Devin is slow sometimes because-

Yeah

... I'm like, well, yeah, o- this is very readable and very sensible, but also it is slower than it could be if I just, like... I want a button to just say, like, "Just ramp this up 1,000 next parallel, in parallel and just, like, see what happens," you know? And, like, I, I don't know if that's, like, s- feasible at some point in the future.

Yeah. I, and we've also run experiments internally where we've basically tried to build entire products, um, like true products that we knew we'd eventually ship, but for now let's try to see if we can do it just by purely, like, vibe coding on top of each other, auto merge, no code review at all. And then there's this kind of benchmark of how many weeks can you go f- onto this for-

Yeah

... before you say, "We have to trash this code base-"

Yeah

... "and actually rewrite it from scratch."

From start to finish. Yeah.

Yeah.

What, what'd you find?

I think we found that the state-of-the-art in December was you can probably, like, run this for about two weeks. By the end of those two weeks you'd find that, hey, y- you wanna, like, change the color of a button. Well, it turns out this button is implemented in, like, 10 different places and they, like have-

Yeah, yeah

... all these different variations and, oh, you forgot one of them, and actually it's a slightly different color one, one spot. And you're like, "Okay, this is too much to work with. Let's actually try to do code review at the same time." A- and make sure that we're on top of our software, actually cleaning it up a bit-

Yeah, yeah

... and making sure it's done in a scalable way.

Yeah. I think building on that, um, the idea of, like, you don't have to look at code, I think is generally a bad idea. Uh, and the, the meme that I have for that-

What timeline, all right, is, is... do you think that statement will be true on?

Uh, I think probably for a while it'll be true that you should continue to look at your code. Um, a problem that I see a lot of teams run into that I work with who are embracing kind of AI native, uh, AI first coding, is the meme that I have is that your code base regresses to your worst engineer because that engineer who is, you know, very gung-ho about AI and is not auditing their code, their pattern starts cementing into the code and now the AI is referencing their patterns.

And so now their if else block that, you know, is 20 if else is back and forth, the AI is seeing that as the pattern of how things are done and starts to then exponentially grow this slop. And I find to your point, a pretty good approach to that is having scheduled cleanup, whether by humans or through systems that are looking for duplication.

They then address that. Um, you know, you'll end up with like 12 helpers for how to format a date, and you need to address that because otherwise it will continue to sprawl.

W- within balance, I think it, it's fine to have some duplication, and then sometimes-

Yes

... to have garbage collection, right? Um, yeah, the, what I've been, uh, talking about with a lot of engineering leaders is that you want to be very strict about the boundaries between modules, and it's your job as an architect, as a CTO, whatever, to say like, "Okay, here's the hard contract between you guys and you guys. Whatever you do inside this black box is your business.

You do whatever. But, like, between these guys, let's be, like, really damn clear, and any movement must be signed off by a human or, or me," or, you know, then, and like-

Yeah

... that's, that's that. I don't know if you have any other modifications or advice.

Well, I, I guess generally on the topic of, like, where humans can be useful, um, I found that 'cause, like, some of these, like, really deep infra problems, sometimes just having a human that just has, like, really deep expertise can make a big difference. I've actually seen this come into play when actually building agents. So we, we've had a few friends now, uh, try building their own co- coding agents, and I think one same problem that I recurringly heard a lot of them run into was this problem of like, oh, like, you know, Grep is really slow on our agents' machines.

And so a lot of them, um, I assume because they're using AI and, like, they themselves don't have, like, super deep infra background knowledge, say, "Okay, we're gonna go build our own custom Grep index. It's, it's gonna be really fast," and use that as a way around this problem. When we ran into this problem- About like, uh, you know, maybe like a year and a half ago when we were, uh, in the early days of building Devin, um, we obviously didn't have AI, then we just asked it, like, how-how to do this.

You can just slap a-a new Grep index, so-

What do you mean you hand-coded Devin? What?

Yeah, it's like, can you believe we-we hand-wrote this code? Um, and-and we-we had, uh, our infra people, who are really amazing, they were looking into it, and they're like, "Oh, you know what? We realized that actually the root cause of this problem is actually super simple," but like fi-fine-grained detail, which is that a lot of these virtual machines, actually underlying them, don't use real file systems.

They use these, like, network file systems where things are actually cached over the network, actually in S3. So when you're Grepping, you're actually making network calls-

Yeah

... ev-every time you're doing these things, and that's why Grep is extremely slow on these machines. Um, and so again, goes back to, you know, what-what is all of the crazy infra work that we had to do to actually get these machines working. If you try to do this yourself, you know, there are tons of small details like this.

And so we had to eventually go swap out that network file system.

Yeah, I think there's a write-up about it, right, Silas? The one about the virtual file system.

Oh, that's a whole other thing. The-the-

Oh, that's a different thing

... the Block Diff file storage format, um-

Yeah, yeah, yeah. I'll bring it up

... which is a file system format that we built so that the VMs could be spun up and down very quickly. Basically, the-the intuition behind this is imagine you have, like, a terabyte of disk and your agent only, like, wrote like a hundred lines of code on top of that disk. How long does it say, take to, like, save and, uh, re-bring up that disk?

Um, and most systems, because you're not optimizing for this case, uh, it's just, like, on the order of a terabyte of work because you have to-

Yeah

... like, save all of that and bring it back up.

Yeah.

In our system, we try to build a file system that incrementally builds on top of each other. So every time you save and bring the machine back up, you're-you're only doing work that is proportional to effectively the diff in the file system.

Yeah.

Um, and so this, you know, shaves off a lot of time in-in the boot up process of Devin. I think we... This is actually now outdated. We have a newer system inside of Devin. But yeah, there's a lot of tiny details you have to get right here to actually get the day-to-day experience of Devin to be good.

It's, like, not technically agents, but it is agent infra, and when you sell an agent as a company, you sell agent plus agent infra.

Yeah, um, at least the way we do it be... And-and the other... The nice thing about having the agent, agent infra being done together is, you know, you... We kind of get to deploy Devin in whatever environment we want now. We don't need to wait for some underlying infra provider to also go and support VPC or on-prem or FedGovCloud, for instance.

So we can actually go and figure out, okay, since we own the infrastructure, how can we get that set up for you?

Yeah. Whereas you're-you're Cloudflare dependent.

Uh, so Cloudflare runs the control plane. Um, the sandboxes modal is supported.

Right.

A contributor just added Daytona. Um, E2B is on the roadmap, and I think there's an abstraction in place that if any contributor wants to add a new provider, they can add that in.

Yeah, yeah.

Yeah.

Um, amazing.

Well, what are, like... How are the customers you work with, do they generally try to then go set up a contract with another one of these third-party providers? Do they try to do the VMs in-house?

Uh, most of them I see using Modal.

Yeah.

I think Modal has a great-

Shout out Modal.

Shout out Modal. Um, I think Modal has a great offering. It kind of captures all of the sandbox pieces you need, snapshots being a pretty big piece of that. And given that they also offer GPUs, um, I think it's a pretty nice offering as a whole.

Yeah.

Uh, no debate there. Um-

No, Modal is great. Especially, I think their-the container offering is, like, the most natural. And so especially if you are willing to, you know, forego, like, the full VM requirements-

Yes

... Modal is, like, a really vast place you can kind of spin something up on.

Yeah.

Yeah.

Is there a point... So Modal's very Python, and I feel like most workload, like, has really shifted to JavaScript. I don't know if you guys get the same feeling.

Mm.

Like, so-so, okay, when I started Lensspace and AIE and all these things, I was like 50/50 Python a-and JS, right? Like, that's roughly. I think that's wrong now. I think JS has won. I-I don't know if you guys... Like, I-I... Maybe I'm overstating it, and maybe for cognition, you know, there's like C# and Java and what have you.

But, like, for, like, new f-greenfield apps, do you feel that... Do you get that sense? Does it matter?

I think that most of the libraries that I see in this space are Python native first, especially in the-

Oh, yeah

... observability space.

Yeah.

Um, that said, I think that there is a pretty big appeal of having your entire system in one language.

Yeah.

Especially when you have both your front end and back end communicating, you can have one central type-

Yeah

... which is very nice.

Yeah. That's my case against Modal, which is-

Yeah

... and then you have to run JS.

Yes.

I mean, you can run JS inside Modal, it's just, like, one extra step that, like-

Yes

... isn't native to the runtime. Um, I don't know if-

Yeah, um-

I don't know if you've used... Do you have numbers? I don't know.

You know, the-the one thing I don't like about Python is whenever... AI, whenever it writes Python, it always does, like, the weirdest patterns, uh-

Oh, because it's, like, mixing two and three or what?

Yeah, I think it's m- it's something mixing two and three. Yeah, like the... I don't know if you see this. It always tries to do, like, has attribute on objects as like-

Yeah, right

Oh my God.

But it's like-

Sure, yeah

... but that you shouldn't be doing that. Like, it should error if it-

Because it's training on library code?

I think it's more of like-

Different theories

... from what I've seen, it's more of, like, a reward hacking mechanism where it doesn't want to basically-

Don't ever error

... yeah, it doesn't want the code to fail. And so it, even when it knows it has the attribute, it'll call getattr on a, um... And for a lot of my clients who have moved towards more kind of autonomous coding, we've put that in as a lint rule-

Yeah

... that if you do getattr, your pull request is going to fail.

Ooh, this is a fun topic. Can you tell me more like this? Like, what else is AI, uh, is, like, a sign of AI, uh, coding that you have to put guards in?

So we were talking just before this about Opus 4.7. One of the things this new model likes to do is it writes lots of comments. Not like, you know, it'll, like, comment every line, but it'll write, like, paragraph, like PRDs, like, you know, on top of every function. Um, but I will say- To its credit, these aren't slop, you know, descriptions like they were before.

Okay.

Like, "Oh, here's what this function does." It's like, "Oh, here's actually the reasoning and why we chose this approach and what the alternatives were and why we shouldn't do those alternatives."

Mm-hmm.

Like, still too much information, but I wonder if this actually might be directionally correct if you want systems that can self-maintain themselves-

Mm

... in the long run.

Oh, they write their specs inline.

Like, have all the context-

Yeah

... in the code as well. Yeah.

So you approve?

I- But at the same time, it's this tricky problem. Like, maybe we'll just give our users, like, a setting or something, uh, for, like, how verbose you want it to be. I, I haven't loved it. Like, I'll ask to... I just... I like the comment, but please, like, get rid of it.

Yeah, yeah, yeah.

But I could, I could see a world where maybe something of the sort becomes reality. I don't know-

Mm

... if you guys know about GitAI.

Yes.

Yeah, so-

We've talked about it, yeah.

GitAI, the idea behind it is-

I'll bring it up, yeah

... that if you run an agent, the actual prompts you send to the agent should be stored alongside the code inside the Git metadata-

Mm

... so that future agents can reference it, maybe code review bots can reference it. Um, and it, it's kind of ideal world where, you know, your context for why decisions were made constantly lives aside, beside your code. And so it's like maybe a more hidden version of this, like, write massive PRDs for every comment sort of approach.

Yeah. I'm waiting for the real bull case where we just get rid of Git altogether. I... We're not... I'm not, I'm not there yet, but I'm looking for it because that would be a big shift.

Kind of on the topic of, like, visible slop, a pattern that I see a lot across GPT models specifically is backwards compatibility, um, at all costs-

Loves, yeah

... where it's doing these weird import exports so that it doesn't have to modify, uh, the names of where the modules were. And, uh, I've seen Claude 4.6 starting to do this as well.

Oh, no.

And again, I think it kind of is this, like, reward-hacking behavior where it doesn't want failure to occur, and you can address that through, like, Semgrep or other tools where that behavior is pretty easy to identify. Um, but it's something that you kind of only learn through the trade of just seeing code patterns.

Yeah.

Um, untyped tuples are a really big problem of just, like, again, just throw any in there, like dict string any. And again, you can address those through linting.

Awesome. Yeah. A- a- any other... So, like, linting, any other tools? Devin review, of course. Uh, not so, not so free now, but, you know, still use it.

The one thing that I think we try to recommend teams as they use more AI agents, it goes back to this, like, local testing thing. In the end of the day, you want your agent to be able to do the full thing, not just write the code, but actually run it and test it. And a lot of code bases were not necessarily built for this from the start.

For example, you probably do want a local DB set up and local Docker Compose and, and Postgres in order to have it so that you don't need to give your agent any crazy product credentials to actually run and test its code. We've also internally done a big shift to make a lot of our core, uh, p- components of code testable as purely local dev without needing to actually, like, integrate with, uh, any live services for this reason.

And honestly, the older the company, like, the more you have to change to shift in this direction. But, you know, you can use AI to help you perform this migration process.

The older, the older the company, the more you have to change in order to do local dev?

I think so.

Or am I misunderstanding? So, so you're saying-

Or oftentimes-

... most, most people just build with full integration to all their stuff, and there's no code path to switch it to local.

Especially in, like... When, when there's, like, lots of different services and you have, like, microservice architecture, making that shift, the larger the code base, uh, the, the harder it is. I guess if you did build it correctly from the very start, that could be possible. But also, like, a lot... There are a lot of companies in the world that got started before Docker was a thing, and so-

Yeah, yeah, yeah

... you're kind of forced to make a migration at some point.

Well, Devin's good, very good at making mock servers.

Yes.

Right? So, uh... And no, the, the... Well, one of the projects that I really want to... It's like, like, it's like Little Snitch. I don't know if you guys have heard of this.

Yeah, I run Little Snitch on my computer.

It's just, like, it, there's, like, a man in the middle, but it, it, like, shows you all the traffic going back and forth.

Yep.

But then from there you can sort of reconstruct the server, right? And then, and then, like, create local mocks so you can local mock everything if you just observe traffic for a little bit.

Yep.

That's an interesting idea.

Um, cool. Uh, I, I, I don't know if this will get anywhere, but I wanted to maybe talk a little bit about the CloudCode, um, uh-

Mm

... leak because usually if I have an Anthropic person on- ... uh, I can't talk about the CloudCode leak. Did you guys learn anything from CloudCode? Uh, I, I...

So, so if I say-

This came out-

... I was not that-

Yeah

... uh, interested in the-

Yeah, yeah

... leak. We didn't spend that much time on it-

Got it

... if I'm honest with you.

I'm just, I'm just, uh, pu- fishing for...

Uh, no, I didn't really-

Oh, all right

... uh, research too much into it.

Fair enough. Okay, one more last thing before we go. Uh, Windsurf 2.0, you guys shipped another thing. So-

Yeah

... the sort of meta context is you use background agents enough, sometimes you're gonna want to bring them to foreground.

Sure.

And, like, that, that little, like, handoff from local to cloud is hard to work on. Um, and then... And Devin has... Or Cognition has just done it.

Yeah. I, I think for me the biggest, um, gap this is trying to close is, again, how do you make the testing process as fast as possible? When it can test on its own and send you a video, it's freaking magical. Sometimes there are just really difficult things you can... that you do just need to, like, pull down locally.

Yeah.

To test. And, you know, we just want Windsurf to just kind of be your, like, local command center of all your agents, like, your, your background ones, your local ones, and you can imagine, "Oh, okay, this agent needs me to review something. I'll pull that down, move my other agents to the background, go test it. Okay, boom, done.

On to the next one," right? Um, you have some issue you gotta fix in the background, just click, like, approve. Okay, set up start a background agent to go fix it. I'd love a world where I don't have to leave this window. You know, then maybe the other window I gotta figure out how to stop spending so much time in is Slack, but maybe, you know, someday-

Yeah

... we'll want to get those tools all.

Yeah. And, like, does that require the binaries to be exactly the same? For local versus cloud.

So the funny thing here is that the behavior between local agents and cloud agents, I think that is actually a bit different-

Yeah

... in their ideal state. I think local agents, you want them to be a bit more fast and let the user make the call on things. Actually, don't try to autonomously go test things. The background agent mode where you go start it off, I think the agent should just assume the next message I send a user should just have everything that the user needs from me, and, you know, not run and stop.

Keep running and don't stop until you have the testing-

Okay

... until you have the full report.

So that's a, that's just a slightly different prompt.

Yes. But for many reasons, because of all the work we do to make sure that Devin works with different Git providers, uh, that it works with different OS' and VMs, like we want as much of that logic to be shared as possible. So for our, our own practical purposes, we try to share as much of it as possible.

Yeah. Yeah. Um, I, I mean, I, I can't imagine how much work it is to, you know, transition back and forth, so congrats on shipping this.

Yeah. Thank you.

Um, okay. Uh, anything else that we should cover before we, uh, wrap? Um, just whatever you guys were talking about in your lunch .

Um, maybe like use cases. Like what are your, like, do, do you find to be like the biggest things that your, your clients are trying to do with their cloud agents today?

Do you wanna just ask it again so we can get like a clean cut ?

Yeah. Yeah.

Yeah.

Because he was drinking his water, yeah.

Yeah, the thing I wanted to talk about was use cases. What do you think are the main things that your clients come to you today about, "Hey, this is why we want to go set up cloud agents"?

Yeah. I think the easiest and most common use case I see across everyone is SRE use cases.

Mm.

Uh, the idea that whether we have our alerts in Slack or Datadog or wherever they're going, we want the agent to be the first responder on that. And that doesn't necessarily mean that the agent is actually resolving the issue, but just being able to collect that context ahead of time is huge. Because again, that agent is integrated into the production logs, the database.

It has full visibility, and over time, playbooks as well for how to address certain issues. And so that's a huge win for teams because instantly you can have a full trajectory of what is going on within the system, and oftentimes actually a pull request directly from that, which is a pretty neat flow to actually experience of like error pull request done.

Um, OpenInspect does support a trigger for that as well, so that could happen completely autonomously.

From Datadog specifically, or just-

Uh, it supports Sentry, it supports a generic webhook, and, uh, if someone wants to add Datadog, they can.

Yeah.

The other use cases that I see, um, are for kind of non-builder use cases, whether that's the PM or the marketing team. I'm seeing a lot of, uh, teams where the idea of who's actually contributing code is starting to change. And in a lot of cases, uh, the PM, if there's just a quick bug fix, the PM is not creating an issue anymore.

The PM is just prompting through Slack, and the pull request is then being created. And so I think that that's a huge win. I, I think that that trend will continue, where we're seeing, uh, code modifications happening outside of engineering. The last common use case that I see is customer support.

Yeah.

And so where they're experiencing an issue with a customer, they're not entirely sure why this behavior is happening. Previously that world was, "Hey, there's a bug when they tried to use this feature. We don't know what's going on." Well, they're now tagging that in Slack. Again, that entire full context is ready. They can then just tag in engineering and have a complete understanding of that issue and completely bypass kind of the, the previous pain points of like, "Oh, can you get more information from them?"

Hmm.

The only things I'd add on top of that I think I've seen is like continual security scanning, continual-

Yes

... security review-

Yeah

... is a very big one as well. The SRE use case, internally we think about it as auto triage-

Sure

... because we just want every message that comes in, and that's an alert, that's a bug report, to have Devin just start triaging it before anything else. And, um, we've leaned into this use case so much so that we've basically tried to make it so that you don't ever have to leave Slack to interact with this. So again, making the interactions with Devin super fluid from the moment the report comes in to it res- it responds to the report and be able to ask it questions right there with full code base context about all the issues.

Very related to customer support as well, I think one thing that we found is CLIs can sometimes be like very difficult for people who aren't technical to go and use.

Yeah.

But, uh, you know, an online chat interface that anyone can go and ask questions and is super intuitive and doesn't assume you have any technical knowledge, but does have access to all parts of your code base, super useful for support-

Yeah

... for salespeople, um, anyone who might need to a- have their questions answered about the code base. So yeah, great callout.

This might potentially be like a very expensive, uh, use case.

Yeah.

Is there like a rule, sense, a rule of thumb on like how much people should spend on this? 'Cause, uh, you have unlimited budget, but like not other people don't. You know, like I don't know if this is an answerable question because obviously it depends on like a lot of factors.

Yeah.

But I guess like-

I think it depends really on, um, how people are using it. I think-

Yeah

... if people are using it responsibly and they're getting value from it, then, you know, you can kinda determine the budget. Common numbers that I hear are anywhere from 1,000 an engineer up to 5,000 an engineer.

Yeah.

I have not heard anywhere in the realm of like 50,000 an engineer for a frame of reference.

We, we'll get there. Yeah.

Yeah. I, I've seen. I, I've seen numbers go that, that high for sure.

Yeah, yeah.

Um, I think that this is also, I think gonna be a big theme of the coming year, is we're gonna see very expensive, very smart frontier models, and we're also gonna fe- see people who say, "You know what? I don't need the frontier anymore for a lot of the work I do," because some frontier models actually are good enough-

Yeah

... for a lot of the work. Um-

Also shout out you pioneered Smartfront, which-

Yeah

... which is a mix.

I'm, I'm really interested in a world where you basically have hybrid frontier and subfrontier systems-

Yeah

... where you use the subfrontier s- part to be really, really fast, really efficient, and call out to the frontier part of the system so that you can still get frontier performance for the most part.

Yeah. I'm trying to search, but Twitter search is, like, completely broken. Like, I, it's like the, the from field is just completely gone. It's very sad, uh-

Yeah

... because I, I really want to-

No worries. Um, I, I, I might have to make a, a, a, a new post at some point about the, the return of smart friend.

Yeah, yeah. I mean, uh, Anthropic has now officially adopted it.

Yes.

Um, okay, cool. Um, I think that's it. Like, it's really great discussion and good, re- great having you guys on. Uh, background agents are a thing now.

Yeah.

And, uh, everyone's building them. Uh, we, but we k- talked a lot about, like, the, uh, the production concerns and like, w- you know, why you would want to offer one architecture over the other. Um, yeah, lots, lots to look forward to.

Yeah. There's a real zeitgeist in the space right now-

Yeah

... I think, for companies to want to turn themselves into these autonomous coding factories. And, uh, yeah, we, we're, we're doing a lot to try to support that. And so, you know, um, any listeners are welcome to, to come chat to us about that, whether using Devin or, you know, working with us.

Yeah. Hiring?

Yes, of course.

Uh, what, what, uh, specifically, you know, just like give, give like one profile that's, like, very interesting.

I think people underestimate the, the role of, like, really high-taste product engineers-

Okay

... in this space right now.

Yeah.

Um, yeah.

And the test is, like, what have you shipped end-to-end that is-

Yeah

... a tasteful product.

If you've shipped stuff that you think is tasteful and you're, and you're proud of, you know, you should, you should come talk to us.

Yeah. For me, um, any businesses that are looking to further their engineering org, a lot of the consulting I do is around that. Teams who are maybe starting their AI journey, whether that's with Cursor or Claude Code, um, but they're looking for someone to kind of help navigate them through the state-of-the-art and beyond just that initial deployment.

Um, as mentioned, there's a lot of lift from you've deployed the background agent to how do we actually get this fully integrated into the company and really realizing the true value of that.

Yeah. Okay. Well, thanks you guys for coming on.

Cool. Thanks for having us.

Yeah, thank you.