← Latent Space: The AI Engineer Podcast![⚡ [AIE CODE Preview] Inside Google Labs: Building The Gemini Coding Agent — Jed Borovik, Jules — Latent Space: The AI Engineer Podcast cover](https://substackcdn.com/feed/podcast/1084089/post/186621812/38ada377118ba1e7fe213ef5b78f06f3.jpg)
Latent Space: The AI Engineer Podcast
⚡ [AIE CODE Preview] Inside Google Labs: Building The Gemini Coding Agent — Jed Borovik, Jules
00:43:53
Loading summary
Transcript152 lines
- [00:05]
A
Okay, Jed Borovic, welcome to lanespace.
- [00:07]
B
Yeah, thanks for having me.
- [00:09]
A
So we're sitting here@f.ins beautiful podcast studios, and we're actually meeting at GitHub Universe. How's it been so far?
- [00:15]
B
It's been great. I mean, yeah, the keynote today was awesome. It was fun to see Jules up there a little bit. We have a lot of folks from our team here. Jules is the app partnering with GitHub for the new Agent HQ stuff, which we're excited about. And also this is an incredible podcast space. So, yeah, I'm excited to do this here.
- [00:32]
A
I'm glad for them to loan us this space. You're also an emcee for AI Engineer Code. That's exciting. In New York, where you went to college but you don't live there anymore.
- [00:43]
B
Yeah, no, I spent a bunch of time in New York. It's funny being part of the New York tech scene. I actually think it's great having big major conferences. There's a lot obviously that happens on the west coast, but being someone in tech on the east coast, it's just awesome to have stuff there.
- [00:58]
A
You mentioned you fly over to SF a lot and like. Like, what's it. What's the scene like in the East Coast? Like, like, obviously we are pretty new. We're like, this is our first year coming to New York. What else happens in the New York. Like, what are the highlights for you in the New York tech scene?
- [01:14]
B
Yeah, I mean, there's so much. There's, you know, obviously a ton of great companies. I think the thing that's interesting about New York is it's such a big city with so much going on. Right. And so there's like, you know, tech is a huge part of it, but there's also, you know, so many major industries there, whether it's fashion, media, fashion, finance. Like, it's like, I think that helps push the tech and do. Do all kinds of stuff. But yeah, no, the east coast is a great city. The great, you know, the, the. All the schools there's, you know, all across the east coast, ton of great schools and great students doing all kinds of stuff. So, yeah, you know, I went to school there. The hackathon scene there was amazing. Really fell in love with tech and. And programming there.
- [01:50]
A
So is there a big NYU hackathon like, like in Stanford with like, Calhacks and stuff?
- [01:56]
B
Yeah, so there's a tree hacks. Yeah, there's one that was put on by. This was a while ago, but we put on by NYU in Columbia we do together Hack and Y. So there's a bunch of events kind of that we did together. It would bring, you know, people across New York City, students across New York City, and those are super fun. Yes. It'd be at Columbia one one time, then NYU the next and just cycle back and forth. So, yeah, a lot of cool stuff was made there.
- [02:18]
A
Nice. So you've been at Google for a while. Nine years. You worked on a bunch of things, including with Malta, which. Which is also another guest that I'm interviewing today. How do you get into Juuls? What's the AI journey?
- [02:31]
B
Yeah, this is going to sound really cheesy, but I've told the story a couple times to folks when they're like, oh, how'd you end up doing this? But it is actually very true. So I worked on Search for a long time and specifically kind of like news and freshness. And then when stable diffusion came out, that, to me was the first gen AI mode. I know some people talk about ChatGPT as the first thing, but for me, stable diffusion, it was a couple months before ChatGPT came out. It was a huge thing, and I was following it a ton online. And there were two groups of creators having reactions to it. There was one group that was, this is stealing my art. This is stealing everything that's near and dear to me. I hate this. This is ruining my life. And there was another group of artists and creators who were like, oh, this is a tool to create better art. And so I was watching this.
- [03:17]
A
A new brush.
- [03:17]
B
Yeah, exactly. And right around then, I was having conversations with a couple people who would say things like, you know, if I had a kid in college, I would. I wouldn't recommend they study computer science. I was like, what? What? And this was, you know, long before, like, Jensen Huang and people, like, he had been saying this kind of stuff. And I was like, whoa, whoa, whoa, why? And it was like, oh, AI. Like, software engineering is going to change. It's going to be so, you know, who knows if there are going to be jobs? And I was like, I love being a software engineer. I love programming. Like, and I was like, wait, this is my stable diffusion moment. This is either it's going to take my art, my craft. This is a tool to create better art. And I was like, I definitely know which path I'm taking. So I got, you know, very into building coding. So I was still working on Search, but I spent a bunch of time making stuff for my own time and playing with things and ultimately tried to find a Role that would. The most exciting role I could find to do this stuff and that was to join Google Labs and Juuls where we were right around then we're starting to build these kind of coding agents at Google. And yeah, the timing worked out well and I joined and yeah, it's been awesome.
- [04:19]
A
Since we're talking about Google Labs, I am actually unclear about where Google Labs starts and the rest of DeepMind and the rest of Google. What is the org chart layer?
- [04:28]
B
Yeah, yeah, yeah, yeah, that's a great question. So Labs mission is to build kind of new like innovative products that the rest of Google isn't well positioned for.
- [04:39]
A
Yeah, which we've had like rise up from Nopaka lm.
- [04:42]
B
Exactly, exactly. OPACOM is maybe the most wide, the mostly.
- [04:45]
A
And then it's called Nano Banana. I don't know if it's.
- [04:48]
B
Yeah, so some of the. So the thing about that's really exciting about Labs is we work incredibly closely with DeepMind. Right. So all the stuff in terms of the, you know, we're building a product, but we work so closely mostly for the model. And one of the nice things about being at Google is you have this opportunity to really build an end to end AI product from pixels on the page through the infrastructure, through the model and the training and all of that loop. So Labs is here to build new products and we're really like a product org, but a true AI product org where we work incredibly closely with DeepMind but also other parts of Google as it makes sense.
- [05:23]
A
Yeah, just on the history of AI coding, I had heard that actually Google had an internal version of Copilot or something like that that was never released. Is that true? What can we say about it?
- [05:36]
B
Yeah, so, you know, I think there are, there are, you know, Google's published papers in this space for a while and so yeah, we have, you know, in Google we built a lot of our own tools and you know, cider, which you know, folks maybe have heard of is our, you know, our internal ide and we've had all kinds of, you know, capabilities and tools there for a while. So yes, you know, we certainly have had pretty good tools for a while, but they were for internal use.
- [05:58]
A
Yeah, yeah, I think, I think it was interesting because like, I think one of the hype moments when Google started getting into the sort of like LM game, like basically when everything rebranded to become Gemini and like starting to start Gemini, where people are like, oh, did you know that Google probably like Google's entire repo is probably about the same size as GitHub and like, you know, there must be some interesting data in there.
- [06:25]
B
Oh, yeah. I mean, and that's one of the things, you know, in building a lot of these internal systems, the data is incredible. Yeah. Especially when it's, you know, not only is the model and the training in house, but all the data around the usage and whatever. So, you know, we could build really kind of sophisticated things there.
- [06:41]
A
Yeah. Okay, so let's introduce people to Jules on your website says Jules, Autonomous coding agents. We've seen lots of these. They're not octopuses, they're not purple. So you got that going for you. But what really is the core thing you're trying to nail in a very crowded coding agent landscape.
- [06:59]
B
Yeah. So what we think about and what we set out to do back when I joined was like, where are coding agents going to go? And as these models get more and more powerful and sophisticated, what is that experience going to be? And let's build for that future. And so when you think of a really powerful agent that can run for a really long time doing really complicated things, that's when, like, the products started to take shape for us. So, for example, autonomous, you know, means like it has its own computer. Right. So for Juuls, it's, you know. Exactly. So tons of agents that run, you know, locally or in your workspace with you while you're coding. But if you want something that's going to run for hours or let's say days, you know, you might want it to have its own environment where it can. Can do its own work. So that's just one of the pieces that's important for kind of this autonomous coding. But it's really like, you know, think about this future where they're incredibly powerful. You can spin up tons of them.
- [07:48]
A
Right.
- [07:49]
B
They're. They're autonomous. But also, you know, we. We're thinking about what is, what does it mean for it to be ambient. Right.
- [07:54]
A
Like, it's.
- [07:55]
B
It's kind of when it has its own infrastructure, its own computer, and its own ways to interact with it, how does that start to change what it can do? For example, like, we have an API, so people are using it for all kinds of things, triggering it from, you know, when something happens. And, you know, we saw an example where someone has. They're triggering Jules for. To do kind of all kinds of updates to their site, and then they have a GitHub that is going to automatically merge Juul's pull requests. So it's just like all kinds of stuff is flowing. You really kind of changing how people are able to do stuff.
- [08:25]
A
Is it CLI related just to close that loop?
- [08:27]
B
Yeah. CLI says we also have a cli, which is, you know, we want to meet developers where they are.
- [08:31]
A
Right.
- [08:32]
B
And so part of the, you know, an API is like you can trigger from anywhere, but also, you know, when you're working locally, like, you want to be able to trigger stuff. So we. Yep. So we have, we have the Joule CLI we launched a couple weeks ago, which lets you interact with it. By the time this podcast comes out, we'll be integrated with the Gemini cli.
- [08:51]
A
So I was thinking, like, you have a number of CLIs, I'm not sure exactly.
- [08:54]
B
Okay, so Gemini CLI, all kinds of places where we're going to kind of mix and be able to harness this power. Right. Because developers work in all kinds of spots and so making it easy to have this autonomous ambient agent that can really do all kinds of work for you.
- [09:12]
A
Yeah. What is your journey? When you started out, did you find any assumptions that were quickly challenged when working with Genai and coding agents in general? I guess you're maybe not too unfamiliar with it because Search uses a lot of machine learned black boxy type things, including bert, which was a major update a few years ago. Yeah. So, I mean, just fill us in. What is your AI engineering journey?
- [09:40]
B
Yeah, totally. So I think one of the things that keeps coming up is like the model makes such a difference. I mean, maybe it sounds obvious, but it's like the quality of the model really changes what you're able to do and how you engineer around it. So, for example, when we started this was with relatively early models of Gemini. We had the agent scaffolding around it was incredibly complex. I think one of the things we've seen is scaffolds get simpler and simpler over time as the models get better. And in some ways the scaffolding is almost a crutch for things the model struggles with. For example, really complicated subagent systems we've played with that, we've experimented with that.
- [10:22]
A
Can you give an example of a kind of subagent that you had to abandon?
- [10:26]
B
Yeah, it was just basically you have. You give Jules a coding task to do and it's going to have different agents for whether it's making a code edit or handling a sub problem or doing any kind of action with an integration, or having full sub agents for different parts of it, like a reviewer agent or even sometimes people do these different Personas where you're like one of the Things that cracked me up is you're the product manager agent and then you have the code reviewer agent. We didn't go, you know, that far. But I think a lot of these things aren't as in favor now. I mean certainly people do, you know, I don't want to say the agent harness isn't sophisticated. It certainly is. But you know, as the models get better, like less is more. Especially as it comes to like being able to improve through whether it's machine learning or just, you know, regular maintenance. I think certainly we found that, you know, we're finding that less is more. I think that you know, that we were talking about a little bit before we started recording like the like rag.
- [11:31]
A
Right.
- [11:31]
B
And like you know, co indexing and all that stuff and you know, it seems like, you know, not just for Jules but kind of across, across the industry that like agent based search. Right. Like it's, you know, maintaining embeddings is hard. Getting the chunking right is hard. Like in terms of like the black box aspect you mentioned, a lot of that is, is hard to improve upon. Yeah.
- [11:56]
A
I would even say it's maybe not even hard in so much as it will never be good.
- [12:03]
B
Well, tell me more. Why do you say that never?
- [12:05]
A
Because a chunk that happens to capture the thing you're looking for will fail to capture something else. And so if you only retrieve based on your embeddings of a chunk, it's uses very arbitrary boundaries that are drawn with some hope of the semantics being captured. But you could just throw attention at it and you can scale probably much better using grep.
- [12:36]
B
Totally. So I think that's an example of these harnesses, how they're simplifying.
- [12:42]
A
Yeah, well, I haven't abandoned it completely because one of the things that we were doing, I don't know if you saw the cognition sweetgrep work was basically using semantic search or chunks with embeddings as a tool, but on the same level as the other tools like code rep and file access and glop and whatever else other variants you have. So I think yeah, I mean that makes sense. Don't abandon it. Just don't reify it into the only way to do things.
- [13:12]
B
And to be clear, this is an area of research we're doing tons of work on and I actually expect, you know, in the coming months we'll, we'll be talking about some stuff we're doing here too. But it's, it's. Yeah, it's, it's not the. I feel like when we started it was like rag. It was like embedding based. Rag. Yeah, was like the thing everyone did and it's interesting to see how it's changed.
- [13:33]
A
People ask me for like, where are the go code embedding models? And you know, I pointed them to like a few, like Chinese ones. There are some like Nomic was working on one and then like we found that we didn't need them.
- [13:43]
B
Yeah, exactly, exactly.
- [13:45]
A
Very bitter lesson. So, so, so, you know, I think like that these, these like, these are good things. Like, I think like when out, it was kind of a preview. I'm in the trusted testers group, so I got to see a little bit, but now it feels like more of a real product. What's that transition like? Is there a process within Google Labs to promote things when you feel like there's some traction?
- [14:05]
B
Yeah, absolutely. So I think the Google labs is not about just experiments. Right. So like notebook gum as we talk.
- [14:13]
A
It's not very serious. It's incredibly successful money.
- [14:16]
B
That's really. And for us, IO was kind of a little bit of a turning point. So in, in May when we, we announced Jules, you know, it was like great reception following I.O. and that was, that was a real moment of us to like turn this into, you know, a very much a real thing. I mean it's something that we were, you know, we always intended to. It wasn't ever intended. You know, I didn't, you know, talk about my journey like I didn't intend to. It was always a goal to build a real product here. But for us, that was kind of a very key moment, very key milestone for us. And so, yeah, now it's very much a real thing as mentioning talking before Jules being talked about in the GitHub keynote. It's certainly here to stay. We're excited to keep building and expanding.
- [15:05]
A
Awesome. Let's talk about just coding engines in general. You're coming to MCD AIE Code Summit. It's going to be your first time at EIE and the mc. What do you want to know?
- [15:15]
B
Yeah, yeah, yeah. Well, sell me. Why? Why would someone want to do it? Yeah, this is. Yeah, let's turn it around.
- [15:22]
A
Oh boy, this is embarrassing. So, I mean, you know, fortunately we're in our third year, fourth year now and we have a bunch of, you know, prior art we can just point people to and say, look at our YouTube. You like that? You like this?
- [15:37]
B
There's some great talks. You know, I haven't been before, but.
- [15:38]
A
I've watched the talk. Yeah, yeah, there's a Lot of good stuff. Yeah. And I'm proud, proud that it features content from all labs. And basically this is a pattern I've seen across my career in terms of every industry needs its focal gathering points to just trade tips and stuff. So I've seen that in JavaScript, I've seen that in cloud native, I've seen that in data engineering. And I was like, probably AI engineering will need something like this. And then the concurrent thread to this was I went to a bunch of the academic ML conferences, NeurIPS, ICMI, clear, and a lot of them, like near earth is 40 years old and that hasn't really changed and is very focused on academics and PhD students. Whereas I think really the transition in AI going from research to industry is that you gradually see a shift, unfortunately. Less open source, less papers and more products and more startups and closed models and what have you. But people still want to share, people still want to hire, they want to promote their work. So they need a place to do that. You can always do that at your company conferences. Obviously IO and GitHub has GitHub and Microsoft has Build and Ignite, but there usually is one place where it's the industry neutral thing where everyone is on the same playing field and me the best person with. And honestly, some people like that. It's not like you're not going to be treated as the VIP and you kind of have to earn your spot, but when you earn your spot, I think people give that the requisite level of attention better because you had to.
- [17:14]
B
Yeah, of course. You know, I've watched the videos online. I kind of get a sense for the speed pick, but what's happening between that for someone who hasn't been before, like, what goes on other than the talks? Like.
- [17:23]
A
Oh yeah, yeah, A lot of, well, just logistical stuff of like invoicing and like vendor selection and venue selection and like, did you know we have like five different pieces of software to coordinate speaker logistics and booth logistics and Bora's.
- [17:41]
B
An av, an attendee. So I'm gonna go, I'm gonna sit.
- [17:44]
A
Oh, yeah, sorry.
- [17:45]
B
But yeah, what am I gonna. What am I gonna get? Yeah, yeah.
- [17:47]
A
So actually it's really weird because, like, as I'm the content guy for aie, right, I curate the speakers, I invite them, but I actually know that the content is like the least important part because all of it's filmed and we're gonna edit it and post it for free on YouTube anyway. But the reason you come is because one, you can talk to the speakers, but Also, you can talk to each other. And so like the, the, you know, I always say, like, the hallway track is the most important track.
- [18:10]
B
Yeah, sort of the hallway track. What's your guide? Beginning hallway track?
- [18:15]
A
I don't have as collected a thoughts as I should. 1 I think if you have some prior history of like, what you're interested in and work on. So basically like the best intro to somebody is if they've seen you online before. So they can skip the whole like, who the hell are you? Part and just get into like, hey, I saw you wrote that thing, like, let me talk to you in person about it. Since you're both here. That's way better than like, who are you? What do you do? And that's like a very cold interaction. Ideally, people come warm or they can come with some clear idea of like, here's why I'm here, here's what I'm looking to get out of this. Because I think if you show up with no real intention or if you're in and out for your thing and nothing else, then you don't have the space and the mental energy for the unstructured, serendipitous connections. And the thing about yai, at least in our scale, our size right now, especially for the summits, which is the one that you're going to, everyone had to apply to get in. So usually our first summit we had something like a 10 to one applicant to invite ratio, invited spots ratio. This one's going to be when it went up to 10 to 16 to 20 something. This one's gonna be 23. One out of 23 people who. Yeah, it's a lot, I think. And really we're trying to filter for people who would be speakers at any other conferen, but like, they are the top of the field. They are either founders or honestly enterprise buyers of the best companies you can find in New York, which, you know. And that's another reason for this, our New York conference, which is we're bringing kind of the best of San Francisco or tech to the finance sector. Really. There is a little bit of media, but mostly finance and like. Yeah, that's great. I mean, I think so. What I'm trying to say, I guess is you're there to meet the other people, so make time to meet them. Have a calling card, like, who are you? Like, like a quick like, what? Who are you? What do you do? What, what can you help with? What are you looking for help for? That kind of intro stuff is really good. Going with friends is really Good. Obviously, like we actually offer. We for the World's Fair, we offer bundle discounts. This one I don't think we do, but just reach out if you could need something. But yeah, I mean like, I think like, like the idea of getting immersed in the code agent community is really important and I think maybe the last point I'll bring up is that we themed it for the first time. Right. So you used to be. These are just generalists. Here's the state of AI. The best speakers we can get at any point in time. But now we're really trying to push ourselves to theme, everything. So we have the best people in code, the best people in data sets, the best people in rl. I want to do a macinterp one that'll be fun. Cool. That one I'm thinking will be in London because the people I want to target are in London. But yeah, I think when you do a summit it should be focused. Everyone there should have an agenda of trying to learn what's the state of the art, trying to have off the record conversations with their peers, doing the same thing at the other companies and who knows what could happen? That's the weirdest thing. I organize the thing and I don't even know half the things that go on. Just because my job is to provide the nexus of people to just connect. Last time we were in New York there were 13, maybe 15 side events organized by people. Just like dinners, meetups, whatever, around this, around the summit. And we encourage it. We posted. We just want people to meet up.
- [21:48]
B
Yeah, I was going to ask, is there a whole off menu set of events happening?
- [21:51]
A
How do people have known? They organize it. Honestly, if you're not scared of strangers, you should organize your own. Like a little dinner. We leave all the evenings open. So just like organize a dinner or meetup, focus on your thing. Like we have people doing only voice. So if you want to do voice, great. If you want to do code review agents as a small subset of generalist coding agents, do that and I think you find it. Right. Or you can do AI in finance, AI in bio, whatever this particular sector might be. And I think like that is honestly the highest signal way to get a bunch of people who really resonate with your thing to meet and have high bandwidth conversations.
- [22:36]
B
Yeah, yeah. Are you going to do the autonomous coding agent dinner?
- [22:41]
A
Well, no, my job is to float. Yeah. My job is to handshake, ask how everyone's doing, fight fires. So I tend to just leave myself open until the end. But yeah, it'll be a sprint. It's always a mad rush because then I have to do my own talking. I don't know yet. I think so far. So the last time I did this summit I was talking about how this year had to develop as the year of agents. And it's really played out a lot. Obviously now the trendy thing is to say no, it's not just a year, it's a decade of agents. But like this year I think agents really took off and most people got it right. Like the consensus was correct. You don't have to be too spicy or counter consensus to say like if you worked on an agent you're probably a lot better off. You probably made a lot of progress this year and maybe you can tell me how it feels from the Jules point of things. I didn't see myself at the start of this year joining an agent company and I ended up doing that. But I've gone so agent pill to the point where people come to me with startup ideas for infra companies. They're like, what if we made agent framework so that other people couldn't build agents? And I'm like, why don't you just build agents yourself, bro?
- [23:55]
B
Like there are a lot of frameworks.
- [23:57]
A
Yeah, frameworks and infra companies and all these guys are just like, they're good developers with no conviction whatsoever in what they want to build. They don't know what customer they want. They're just like we want to build developer tools because that's where we feel comfort comfortable. But honestly it's not that hard to actually take a stand and be full stack and verticalize in some particular agent field that you want because guess what, the business and the economics are know align that way. And I'm not saying that you cannot make it as an infra company. There's some fantastic infra companies that are sponsors and like that I admire and you know, I would invest in myself. It's just that comparatively those are a lot harder. And like agent companies seem like they're shooting fish in a barrel. They seem like they're ramping up in ARR a lot faster and it seem like their margins are better. So why not?
- [24:48]
B
Yeah, totally. I mean I think for us it's really been you're the agent like as the models, you know, as you were talking like what is let's build jewels for where things are going. And as the models get better, I think it just becomes clearer and clearer that agents are super powerful. You know, like we have, you were talking about like before high context and management. So all that stuff's important. Like we have people. We had. This is a funny story. We, we store some data for a session, but it only lasts. We only store for 30 days. And so after 30 days your, your session becomes locked and when the first user starts, first start hitting that. That they were upset. We were like, there's no way anyone's going to be using a single session for 30 days. Like maybe we're doing a single track of work for 30 days, but just like how powerful that could be. So. Yeah.
- [25:30]
A
How do you compress context when you run into it?
- [25:33]
B
Yeah, so we have, I mean, I can't talk too much about it, but we, you know, we do a lot of the standard things and there's also, you know, we're developing a bunch of stuff. It's an activated area of research for us.
- [25:42]
A
Yeah, I think like, you know, just, just to. I'm not asking you for how exactly Jules does it. There's just a number of approaches. Right. And you just have to pick one because you can't just use up your 2 million token context window. Is it 2 million? It is up to 2 million especially.
- [25:58]
B
For coding agents because you're reading files, you're running commands with huge outputs. I think coding agents are a really interesting area, both product wise and the impact they're having, but also for research, if they really push the limits of what other domains are you running an agent for 30 days? What other domains are you accumulating? So much context and so many turns. Yeah, coding agents are I think, kind of a special spot of like super interesting product impact research.
- [26:24]
A
Yeah. I see the AMP folks drop the auto compaction for a handoff mechanic which was pioneered by the agents SDK, which is basically the subagent's pattern where like you spin up a subagent and do a thing, you don't need all their context that subagent is doing and then you can sort of come back to the main thread.
- [26:43]
B
Totally. Yep. Yeah, it's a, a good pattern. It also has this challenge is like, how do you make sure enough stuff, information is going back and forth. But that's the pattern, the summarization is a pattern kind of externalizing some of that context, whether it's like writing it to a note kind of thing is a common pattern. So yeah, there's tons of things, tons of things to try and do.
- [27:05]
A
Yeah. And one thing I do want to get more consensus about is what is the best because I don't think I've read any papers, but which Methods compare better.
- [27:15]
B
It's also interesting as the models change, the answers change a little bit too.
- [27:19]
A
Yeah. Claude, you probably know, Claude externalizes too much.
- [27:23]
B
Yeah.
- [27:27]
A
How much does your work? Actually, I feel like I switched back to Jules mode. Yeah.
- [27:31]
B
He flowing here.
- [27:33]
A
Well, I mean, how much does your work inform the model creation? Right. At the end of the day, you obviously are a very big consumer of Gemini models, but also you are not the only consumer and they have other priorities than you.
- [27:45]
B
Yeah, totally, totally. I mean, I think we're lucky in kind of how we're positioned to. We have very close relationships with DeepMind. So we have. And you know, coding agents are an important area, let's be honest. Right. Like for any kind of company building models. Like you can see it in all the labs, like coding agents are important. Coding capabilities are really important.
- [28:04]
A
Yeah. My OG image of the AIE code, I wrote something obnoxious like code is the first spark of AGI, which is like, probably true.
- [28:15]
B
Totally. Yeah. It's important from kind of an AGI perspective. It's important from a dollar's perspective, it's important for all of it. So it's. I think we're in a really lucky position. Yeah. We have, we're able to have a lot of kind of good collaboration, but both ways, you know, like all kinds of capabilities that are being developed.
- [28:33]
A
Yeah.
- [28:34]
B
And you know, it's interesting.
- [28:35]
A
It's.
- [28:35]
B
It's a whole host of things. Right. Because in terms of AGI and the capabilities of things, it's also like computer use models and browser use models and so models that output code. But it's also the whole suite of things that you'd want an intelligent agent to be able to do. So it's multimodal. It's all kinds of stuff that goes into it.
- [28:58]
A
What would you want to find out from your peers at other coding agent companies? Because you're going to meet all of them, basically.
- [29:03]
B
Yeah. So I think one thing, and you know, I don't think of this as a zero sum thing. I think this is like really like there's this tide that's going to lift all of our boats and it's. We're inventing a new way to do our art. Right. You know, and how to create good art as a software engineer. And so what does that look like and how does that feel? What is that? You know, what is the experience we want to create? I think as people working in AI, sometimes we don't do a good enough job describing this beautiful future we're creating. I mean, I know like, you know, like the CEOs and heads of these labs have started like, you know, writing their think pieces on this. But right. You know, for software engineers, like, what is this beautiful future we're creating? And like, you know, I think that's like one. It's inspiring. It makes it, you know, maybe less scary for people who are, who are thinking about these tools, but also like, you know, if we can't articulate it and think about it, it's less likely we'll get there. Right. So, like, what is this great place we want to create? Like, writing software is so hard, like in so many companies. It's such a. Companies, it becomes so challenging to manage a code base and create and, and what can we do to make, you know, being a software engineer an absolutely incredible experience? What are these, you know, how do you want to interact with your model? How do you, how are you doing things locally versus in the cloud? And how does that interop? And so I think like, as, as an industry, we're trying to like, you know, which is change. Like we're inviting in some ways inventing and there's this movement to, you know, change how we do our art. And yeah, the better we can create this experience, we all win to some degree. So, yeah, I think that'd be one.
- [30:36]
A
Thing where it's like, yeah, the local to cloud sync is the most contentious or important, I guess, topic for a lot of people. I wonder if we'll ever get some kind of interop thing. Probably not, but Mac and dream.
- [30:52]
B
Tell me more about what's your dream flow here?
- [30:54]
A
I don't know. Start with Jules Cli end up in dev it. I don't know.
- [30:58]
B
Oh, interoperating agents.
- [31:00]
A
It's probably meaningless. No, but like, I'm not actually serious about it, but like, well, so I think Codex or is it cloud code? Cloud code Web can do this teleport where they just basically dump like the entire history and you can pick it up in cloud code on your desktop and probably that's the right move. Yeah, maybe there's some more sort of elegant things. But they were first, so why not? Actually, maybe the real thing is. Maybe it's not the conversation. Maybe you don't need to teleport if the unit of the artifact that you pass back and forth is the linear ticket or the GitHub PR. Right. So you don't need the full JSON, you don't need the full chat history. You just need to pick up where other people left off because that's how Humans do it it.
- [31:50]
B
Right.
- [31:52]
A
I don't transfer my brain state to you. I just tell you what I did and then if I forgot to say something, you find out eventually.
- [31:58]
B
Right. You say the cloud agent dumps some kind of summary onto the ticket or whatever kind of it needs to pass.
- [32:03]
A
On to the next in Slack or linear and whatever.
- [32:06]
B
Yeah, that's interesting. I think there are some patterns emerging, like ide, cli, Cloud. Right. Like these are the pieces.
- [32:16]
A
Vs Code extension.
- [32:20]
B
Like there's a. Like the surface area is like standardizing. It feels a little bit. And yeah. How these things interop. How you can kind of make this like great experience between all of those. Yeah, I think it's really interesting.
- [32:33]
A
Yeah. Yeah. I think like. And then the other point, I just want to backtrack a little bit to something else you said, which is like what the thick pieces that the CEOs and stuff do. I think there's a lot of question about the impact that coding has on the software engineer industry in general. The humans. Do we end up. Do we stop hiring juniors altogether? Is it actually increasing productivity or do you just feel like you're increasing productivity? I don't know if you have any take on that stuff.
- [33:03]
B
Yeah, it's only. So I mean, something we spend a lot. I spend a lot of time talking and thinking about with folks. I also spend time talking to people at companies and I think sometimes working on these tools. It's interesting to see it's not as like diffused. This technology isn't as diffused across software engineers as I sometimes expect. Right. There's plenty of places that I think are not really using AI a ton. A lot of companies, a lot of software engineers aren't. That being said, I'm very kind of excited about what the future of software engineers are like. Could you imagine going back to not having these tools? No, that sounds horrible. Right? Like that. And so that's one aspect of it. But I also think, you know, I don't really buy this like, you know, that we're not going to hire more software engineers story. I think, like for a few reasons. I mean, this is an example that often comes up. But is it like kind of the elasticity of the demand for software?
- [34:00]
A
Yeah, Jevons paradox.
- [34:01]
B
Exactly. And you know, like a lot of the cases sometimes come up is you look at like farming. Right. And so, you know, there was a time in America where like the vast, vast majority of Americans were farmers. Right. And then technology happens and today it's like less than 1%. Yeah. And that's one example.
- [34:15]
A
But the flip side of that is.
- [34:16]
B
You, you electricity, which like as that gets cheaper and cheaper, people just consume more and more and more electricity. And you know, with food there's only so much food we're going to eat. Right. There's, there's a kind of a, there's an inelastic demand for that, whereas just a very elastic demand it seems like software, you know, software keeps getting better and better. Like the ability, like we're creating more and more software from like, you know, obviously like punch cards through to where we are today is like remarkably different in terms of how you're able to create software. So much more software is being made and software just keeps becoming more and more of our gdp, right. Like it's, it's a. So I'm, I'm, I'm bullish on kind of the, the amount of software we'll be able to create, how it'll be created. I think there's also something here about, you know, as, as an engineer being able to be more productive, like encourages more investment in people building software. Right. If it's, you know, the job of a software engineer can now, you know, they can do 50% more, 100% more, 10x more like justifying investment dollars into projects like dramatically changes. Right. And so yeah, I'm bullish on this idea that is actually going to be great for software engineers, both for our ability to kind of do our craft or art, but also just what it means for the number of companies and the amount that's made and the quality of it and what we're able to do with it. So, so yeah, that's my rose colored glasses take.
- [35:39]
A
Rose colored glasses indeed. Yeah. I have this take on the different kinds of work. Like we're splitting up the different kinds of software work and there's a lot of commoditized work that we used to spend a lot of time on and now we can basically entirely delegate to agents. And then that leaves us ideally for more strategic, important, novel, high risk, whatever work, deep focus work. That is something I posted here on the semi async value of death where basically you kind of need to, on the extreme end you can delegate to async agents which Jules cloud code, whatever. But then over here you kind of need the sort of deep involvement in understanding the code base and not vibe coding, whatever the opposite of it is. Actually that's my talk, which is I've been thinking about this, so I tweeted out this phrase because I feel it's in the air that the term vibe coding was obviously coined by Andrej and he's super influential in February and people have just come to use it as a blank check to just YOLO on prompts and stuff and create the worst code imaginable and leave other people to clean it up. So I think people are kind of at their limits with this. It's probably maxed out in terms of popularity, but we don't have yet. What's next? So my talk is really challenging every attendee, every speaker to come up with what is the aspirational good version of vite coding that we can actually trust?
- [37:12]
B
Yeah, what is it? And it's the punchline right now. What is it?
- [37:17]
A
The current leading candidate is agentic coding, which is what? Dharmesh Shah, who's like, I don't know if you know who Dharmesh is. He's pretty good track history when he's naming things. It's just too many syllables. I don't think it just has the. It doesn't have the joy that vive coding invokes, which I think people want. But then people also want care and craft and reliability and all that stuff that.
- [37:42]
B
But if we don't have the term I describe it, maybe we don't have the precautious phrase for it, but what does it look like even if we don't have the phrase?
- [37:49]
A
Yeah, that's a great question. Well, we have some speakers who are going to be pitching spectrum and development that you have to really be thoughtful and effectively write a prd. I think that is obviously correct in terms of basically it's just a glorified prompt, but a very good one. And models are tuned to follow your prompt for good and for worse. If you prompt sloppily, you're going to get slop. So a spec sounds good, I think. I don't know how often it'll be followed in practice because effectively what that transitions us to is a waterfall development approach where you spend three days writing a 50 page document and then you kick off the agent. That doesn't seem right. So obviously I have some bias here because cognition has from the start believed in interactive planning where you kick off a thing, you get some feedback, then you're like, oh, that's not what I meant. Let me correct myself because I don't know what I wanted when I started. So you work with the machine to discover what you wanted, and the machine works with you to either get you what you wanted or show you the errors of your ways and then you correct it from there.
- [38:59]
B
Yeah. I mean, one thing we talk about which vary on is what you're thinking is like, they're kind of like two problems as these things get better. One is like, how do you specify what you want? And the other one is how do you verify that what you got is what you intend? Yeah, yeah. And so, yeah, whether it's, you know, specifying through a spec or this like, you know, interactive plan or whatever it is. But then. Yeah. And then on the flip side with the vibe coding thing is you might specify, but you're never to come back and verify. Right. You're like, you're. It's more hands off the wheel. Like, maybe I'll click around the app a little bit and see how it works.
- [39:28]
A
But it's.
- [39:29]
B
I'm not really engaged with the code. So how do you. Yeah, how are you verifying and making.
- [39:32]
A
Sure that it's, you know, to my knowledge, you guys don't emphasize tests that much. Right. It's not like you volunteer to write my tests. Yeah.
- [39:40]
B
I mean, it dep. It depends. Like we. If there are tests in your code base, it's right.
- [39:45]
A
It's right out of the picture here. Juuls will run your test feed.
- [39:48]
B
Exactly, exactly.
- [39:49]
A
So, but it's not like. It's not like, you know, after everything, everything must have a matching test. The prompt that was mentioned, that would be the extreme of what we mentioned.
- [39:58]
B
I don't know if people always want that. Maybe it'd be helpful to do that, to kind of show that it was right. But let's say I don't write tests in my code base. I want to merge that pull request that is introducing tests just for this one thing. I think in some ways the engineer should be able to control what kind of outputs they want. If it helps and they want it. Absolutely.
- [40:18]
A
And then do you think there's other innovations on specifying apart from just traffic that.
- [40:22]
B
Oh, totally, totally. I mean, agent.
- [40:25]
A
Agents. Md.
- [40:26]
B
Yeah, agents. I mean, spectrum development, I think is in this. This category. I think one of is like multimodal. Right. Like, you know, if. If I'm going to show you a bug on our website, like, do I want to come and like type it with words to describe it or am I going to point. Yeah, the picture.
- [40:43]
A
Yeah.
- [40:44]
B
And so, you know, with jewels, you can upload images now, but, you know, kind of. Of more, you know, we have certain ways we communicate as humans that are easier in certain situations.
- [40:54]
A
Yeah.
- [40:54]
B
But let's bring that to our engagement with the options.
- [40:57]
A
So of all people, I Expect you guys to be best at this because Gemini has video understanding. I want to submit a video because some things like do cannot be screenshots. It's more about the behavior of things appearing and disappear. Yeah, I mean, I would love that if you guys did it, because no one has it yet.
- [41:15]
B
I know. I would love it too.
- [41:17]
A
I'll tag you over your Nelson on my side. The version of that that we're exploring is computer use. Computer use was kind of introduced by anthropic and then OpenAI did their toe in with operator and now Agent mode in Atlas. I don't know if you guys have done anything super smashy on computer use, but anyway, it's coming back. I can feel it.
- [41:39]
B
Yeah, I think. Yeah, definitely. And it ties into. It ties into coding agents. It ties into just using AI systems in general.
- [41:47]
A
But basically your VM now needs to render a UI or browser and then you need to let the agent click around in it. Absolutely. And you need to have precision and speed and cost and like, you know, affordable cost.
- [41:59]
B
Yep.
- [42:00]
A
It's a lot.
- [42:01]
B
Yeah. What can I spoon? Products are so fun. There's just so much to build. There's so much. You know, I think also as a software engineer working in this space, like, I think one of the reasons we, you know, you see so many companies in this space is probably like, it's just so fun. Like, there's so many things to go. There's so many tools that seem like, you know, fun sci fi. Like, there's. It brings up a demo of what I've worked on. It's clicking around and I can see a video of it or I can even take over and use it, like. So, yeah, it's. Yeah.
- [42:26]
A
Awesome. Okay, so just moving towards wrapping up. If people run into you at aie, they've, you know, they heard your pitch on Jules.
- [42:35]
B
Yeah.
- [42:36]
A
What else should they also talk to you about? Like, you know, what, what. What are you. What can you help with versus what are you looking for?
- [42:43]
B
Anyone should feel free to come up and talk to me, you know, at any point. I, you know, obviously very interested in anyone who's doing stuff with coding agents or someone who's using coding agents in interesting ways. I'm always curious about, you know, workflows people have with their daddy agents. Whereas, you know, whether it's, you know, hey, I'm using this tool in this way and I've, you know, configured this crazy thing. Like, I always love hearing how people are using it. I also love hearing people who are having bad times with it where it's like actually I don't, you know, maybe they're not coming to this conference but you know, I, I Jeff, try all these tools and I don't like them. I don't use them and here's why. Yeah, so, you know, I'm totally open for any side of, of the all the way from, you know, full AI pill and coding, coding AI lovers to people who hate it. As far as what I'm looking for, you know, I think, you know, really just going to kind of connect and meet people. I think you know, we are always hiring so like, you know, I'm I anyone who's, you know, interested in working on this stuff, I'm always happy to talk. But yeah, really just kind of, you know, meeting people, spending time geeking out on this stuff.
- [43:38]
A
Yeah, there'll be lots of geeking out.
- [43:40]
B
Yeah.
- [43:40]
A
All right, thanks for your time. Looking forward.
- [43:42]
B
Yeah, same.