Jori Lallo on Linear Agents, AI coding tools, and why they don't chase AI hype

looking at the new uh Sam Alman IO acquisition. Uh have you do you have a take on the new device? What do you want to see in the world of AI hardware? I I don't know to be honest. Let's see when like things start shipping and like how things like evolve. Uh it's been Have you had a chance to play?

I mean like it's it's great to see like I've again like in you know driver's seat and like on the on the news like doing uh doing interviews and so on. So I think like it's like we had a had a break from for like whatever like four years or like so it's it's been a while. John John called out.

It was like very strategic to go have him do a big interview at Stripe Sessions and then immediately the next week, you know, come in with this massive announcement. Uh but I but I'm excited to see I think uh every every technology Call that a coincidence in my parts. Yeah. Yeah.

Uh but I think every every technology enthusiast will benefit from him taking a real crack at at AI. What what else in the news this week? It's been kind of an informal AI week over here at TBPN between Microsoft Build, Google IO, OpenAI, IO, linear linear agents, linear agents, and anthropic claude 4 dropping.

Uh, what's been most interesting to you? What are you most excited to leverage? Uh, pretty pretty bad week for me for like following the news. It's like finished up moving.

So I've actually been like relatively like off news and like I think like on the docket for like a long weekend to like start catching up more on the on like all the keynotes and so on.

But I don't know it's it's it's good stuff like having more agents or like especally coding agents and like different like players like coming coming into the space.

So, uh I don't like it's it's it will be like interesting seeing like over next couple of weeks kind of like what like performs like how well and also like uh what kind like u user experiences are going to be there.

Uh because I think like that's that's kind of like the bit that we're we're like at linear like now following much more as we're like also you know trying to like work with like different players and like integrate stuff into linear is like what are like the patterns of like usage that will happen uh with this. Yeah.

What's more important just vague definitions around user experience patterns for agents versus standards like MCP and more technical advances. Yeah. I mean like those are like very two different things.

I think like you know like MCP and like open standards like that's like personally I'm like very very like pro of course as an engineer.

Uh but it's been something like I think like we kind of like lost in like last decade of like the big platforms owning more and more on their like own like trying to like hoard everything but like now things are pretty much like being pushed out in the open.

Uh and people are like companies are like almost like forced to cooperate in like the new world. There's just so much demand. Uh and it's been interesting to see from our side too as like large companies like jumping very early and like like a whole space is kind like pushing all the companies to like act fast.

Uh let's see how much is like you know hot air like what like will come out of it but I think like overall it's like push for openness is good.

Um but then how have you so yeah so for for linear's agents product how do you evaluate how how open is the product if if somebody's building an agent can can they build an integration with with linear by themselves or are you guys really kind of uh doing your own internal testing to kind of qualify potential partners there?

Yeah, we're we're kind like looking more on the promotional side, I would say. Like we believe in like open APIs and like trying to like have people access, but then I don't like the uh when it comes to like security and these kind of things, it comes more on like what do you like promote? Sure. Yeah. Yeah.

I mean, at the same time, like if I get a really amazing agent that just can puppeteer my mouse and keyboard, you can't really do anything about that. Maybe you throw up a capture, but agents are gonna be able to solve those pretty quickly.

Uh, and so at a certain point you have to think like like you're you you know there's a reason why you have an API, there's a reason why you have an MCP server, there's a reason why you have maybe an agent integration, but are these like temporary steps in your mind or do you think that there's real durable value to um an agent like a standardization something that you build in house some sort of protocol or some or standardizing against things versus just like MCP I kept coming back to Can't the LLMs, if they're really so smart, can't they just use websites?

But what what's your take on that trend? Yeah, it probably comes like down to like the user experience that you want to like offer. So, you like uh controlling a computer or like VM or whatever. Like that's the kind like the ultimate like bandaid to like everything.

Like it allows to do everything, but it's still like it's not purpose-built. Mhm. But and like MCP to a degree is like look like similar. It's just like a way to like do like do uh um request response like calls and get the information in and like share that.

But then it's more about I think like in our mind it's like what is the experience like inside the product that like you're going to offer that maybe is like specific to a product u and how that does how does that work like today like our implementation is like you know somewhat like rudimentary that we would expect like agents look like regular users but it's not very exciting to see the um the progress of the agent when they're like posting new comments.

in the thread. That's a little bit of a hack uh that you kind like have to like live without while we still like see what's actually like required and like what can we like build to like better like support like the agent developers like what what kind of like experience to like give them.

Um of course like we're you know like looking at this from the lens of like linear uh of course like running the company and like building the product. Uh but it's also like it's it's interesting to see how um the like where does the invocation of like agents happen? Like is it like your CLI is like a text editor?

Is there like some kind like web tool or like the desktop app? Um of course like for us it's like it's pretty natural like being like close to like where you work.

So like integrating into that workflow and I think like that's why a lot of like these like smaller companies especially like are pretty excited to like build in linear because they're already using linear themselves. What what agent uh gets the heaviest usage internally at linear today?

I think it's being like couple of the coding agents but those were like the first ones to you know to the market. No, I I mean I mean not even I mean specifically like your team at linear. Yeah.

Like coding coding agents or or Well, you know, I I kind of like mentioning names because like of course we're we don't have a horse in the race. We're just the interface for this. I love I love horse races and I love having horses in races. Yeah. Pat, pick a horse. I know.

But like it's changing like every week as like you said like there's like a lot of like new new agents that like came out this week from Google and like uh open AI and so on.

So I just kind like it'll be like interesting to see like how those like play out and we're like looking for like working with like all of them of course but I don't like beyond like like currently the focus is a lot on just coding agents.

um like it's it's natural because there's a lot of like value to be created over there but I think like we'll in the next couple of months we'll start seeing much more like non-coding agents like augmenting like in alongside with the coding agent so you might have a you might have a task where you have a you know like a feature like feature flagging like agent coding agent and I don't know like who knows a marketing agent like helping you out like write the change log blog post or and so Yeah.

Yeah, that makes sense. Uh, are there any other AI features that you're implementing that feel like Windsor for cursor for project management?

like something that like lives in linear alongside but it's not that asynchronous because in coding we're seeing a lot of like there's synchronous AI and there's asynchronous AI and these two two different patterns and it feels like the it feels like the market's bifrocating but no investors want to admit that because they want to say it's going to be winner take all but in fact uh we're seeing two different paradigms kind of emerge or maybe even three different paradigms emerge um what are you seeing on the project management side yeah that's Um yeah I don't like the agents are roughly especially in our case are tied to issues or tasks and that's kind of like the interface for them.

Outside that where of course like linear is not only like issue tracking tool it's like for project management like organizing like your your work at the company level not only like individual level so and that's where kind like have like separate work stream of uh we're building like our own like AI tooling around that.

How can we augment the project managers and like the product leaders to like do their work better?

And that you could like imagine looking a little bit more like a a cloud or cursor like alongside your linear data and but like with that it's going to be interesting to see like how we can start like introducing like the external like agents or like MCPS as like part of that.

um that of course like the issues is that's like the first first step because like much more natural.

What's your team's approach to testing new for example coding agents right it's uh in many ways like the linear approach from my perspective is you know really thoughtful ideally long-term planning around building you know beautiful products and and taking a calm approach to doing this and and in in that way you know enabling other people to maybe have more calm uh uh effective uh product development.

Uh, but at the same time, like testing new tools, they can be super effective, but testing new tools can also be super distracting. You know, we don't do a ton of engineering on TBPN.

We actually do a surprising amount um kind of like back office stuff to kind of automate production, but um I I can imagine an environment where you have, you know, call it 50 engineers and and on any given day, one of them can send a message into Slack, hey guys, you got to check this out.

it's like amazing blah blah blah and maybe they just got like a couple good results and it's actually not worth directing everyone's energy to but at the same time you want to stay at the edge but do you have a philosophy or kind of an internal ethos around testing new tools? Um not really.

I think like mostly comes from people themselves. Uh we're not like enforcing like certain tools. We're of course encouraging couple like couple of tools where we can you know like maintain like security and those kind of things. We're like mature company at this point.

So like we need to we need to look after like our customers information and like a lot of like like also lives inside linear. So you know we we need to like look after stuff.

Um but when it comes to like new tools like the the team is like looking at the news the same way we are and like trying out stuff and I think it's a little bit more so you you put out like liners ways a little bit more calm and I think like that also shows on the tool adoption and so on like you you're like excited like to try out stuff but like you know you take it with a grain of salt instead of like going on Twitter and like plasting like we replace all of our team like with AI.

Uh I mean like that's one way to do it, but like that's definitely not us. Yeah, nothing against that either, but yeah, the Clara method. Not naming any I said I said it. I said it. I I respect it. It's good marketing. It's good marketing, but it's not for everyone.

Yeah, I like that that's that's been like overall like when it comes to developing AI tools like we started like when like everyone else started like couple of years back when like the first GPTs like came out and I tried to do stuff and like you know tried to build a chatbot and so on like but I think like very early on like we realize like well you get to like whatever 60 70% but like then it's like falls apart uh the experience and like it's And then it's really like hard to figure out like what's actually path to get to like the 95% where you want.

This is the Apple problem. You you linear and Apple have similar sort of like desires for perfection, right? And and generative AI is in many ways completely imperfect, right? And and unreliable. So that's an interesting it's an interesting challenge. Yeah. But we we built we built a few things. We shipped a few things.

We didn't make like a maybe like the biggest fuss about it. We're not the AI powered like issue tracking software uh raising like gazillion like dollars. Uh but but raising customers though. Yeah. Yeah.

And I mean like in the end like they they want to get their work done and like do they want to like buy hype or do they want to like buy product?

Maybe today like you you want to like buy more like AI and that's where we're seeing like a lot of like demand for AI solutions and like now we're heavily investing in that like and that kind like our we did a course correction over like the last u like roughly 6 months ago when I think like the tools has got so much better like for me personally I was like on parental leave and like when the deepseek like came out and like just trying like the thinking mode was like the light bulb moment like I think interesting.

Yeah. A little bit background on that. I think just because of like linear philosophy is like try to build really like you know snappy like purposeful tools like we always like chase the millisecond and like try to get something like really fast and when it comes to like LMS like you have like inherent latency to it.

So it just felt like how do we how do we bind this like weight into the product that's instant. Yeah.

And I think like now with the the new like like thinking modes and so on like that has changed the like user like um what people expect like that paradigm and now we can like build towards that and like people expect take a little bit like more time uh and also like more UI patterns to support that but like you get a lot of value out of it.

Totally. Well, yeah, check out some of Google's launches from earlier this week. What was it? They were doing 3,000 tokens in like half a second with the diffusion models. Yeah, very cool. Um, anyways, this was awesome. Congratulations uh on the launch this week and uh come back on again soon. Yeah, hopefully. Yeah,