Linear CEO Karri Saarinen: "Our customer base is quite powerful"

There are a handful of tools that almost every software developer on the planet has grudgingly used as part of their day-to-day workflow over the last decade or so, and Atlassian's Jira bug-tracking system ranks near the top of that list in both daily usage and off-the-record complaints. Linear has raised over $50 million to build a modern alternative, and it counts some of today's fastest-growing tech companies as customers.

Founded in 2019 by CEO Karri Saarinen, Tuomas Artman, and Jori Lallo, Linear started off as an issue-tracking tool helping developers coordinate when eliminating blockers and fixing problems. Since then, the veterans of Airbnb, Coinbase, and Uber have expanded the tool into a broader product-development system, which allows designers, product managers, and software engineers to work with their customer-facing counterparts to prioritize roadmaps and triage repairs.

"We have this fairly simple idea that engineering is really the front line of all this information," Saarinen said in a recent interview with Runtime. "If we can build a tool that they're happy to use, they use it more frequently, all of their information is more up to date, and this kind of tool that is supposed to be a system of record is more valid for the companies" that use it, he said.

Linear now has 14,000 customers using its software to build their own products — including OpenAI, Scale AI, and Brex — and unusually for an enterprise tech startup of its age, the company has been profitable since the beginning, Saarinen said. Next week the company plans to launch Linear for Agents, which treat customers' AI agents "as first-class users of Linear; they can be invited to teams and workspaces, receive assignments and take action, just like human users," according to a company representative.

In the interview, Saarinen discussed Linear's role within engineering organizations, the company's initial hesitancy around building generative AI technology into its product, and the tipping point that prompted it to jump on the agentic AI bandwagon.

This interview has been edited and condensed for clarity.

Do you see yourself more as replacing Jira? Some of the companies I've talked to that have used your product have compared it to a replacement for GitHub Projects and Issues.

Saarinen: It's more like we try to be the system for the product team, so that includes product managers, designers and engineers. Engineers are probably the largest use case, but we also have a lot of the product people and management and leadership using it as well.

I think, in reality, engineering organizations don't want to use multiple tools to track work. So in the short term, there can be situations [where] companies might use multiple tools. But in the long term, we think that they should just pick one. What sets us maybe apart from some of these other tools is that we look at this as more like a workflow-based tool.

I come from a design background, so I always believed that form follows the function, not the other way around. Every company has to do some kind of bug triage, but no company has implemented any kind of system for it; usually it means that you have to build some kind of custom thing where there's some kind of backlog and a certain way of managing it.

But we actually built a triage feature that is basically like an inbox queue that you can set to have new stuff automatically going there. You can also set someone to monitor that queue so you can have a rotation based on your on-call responsibilities. And those people get notified [with] specific actions they can take, like, "Hey, we should accept this, we should reroute this, we should cancel this, because this is not actually a bug."

We try to figure out, "what are the main things people actually need to do," and then build the features around it, versus having this kind of endless platform or some kind of product that can do anything, but it's not really designed for anything specifically.

How have generative AI technologies have changed your product development thinking within the company, building up to the Linear for Agents release?

Internally, we've been trying these tools. We've been following the space and trying different experiments with a lot of things, but felt that things weren't quite there yet; you could get some nice, impressive results, but it wasn't that repeatable or that precise. Our tool is used very seriously in organizations, and I think there's a little bit less room for errors or mistakes to happen. So we [thought], we can be maybe a little bit more conservative on the AI front.

But now we see that the models are getting better, and we're thinking about it in two ways. One is that we have this idea of product intelligence. There's this opportunity to make some of these workflows smarter. For example, that triage case; a lot of times, the problem in companies is that you are tracking bugs, but then they should be routed to the right team or the right person, and right now it doesn't happen or it's manual.

We've been testing it out — and we have a good way to potentially do it — with an AI model. Linear could understand your product areas, and then also figure out which teams or people should manage that. Then when the issue comes in, it could kind of automatically [decide], "Hey, I think this issue should go to this team or person."

Similarly, we have this feature around figuring out duplicates, which is also a huge issue. Sometimes customers say that potentially 50% of their backlog is duplicates, but I don't know which [ones]; there's not an easy way to figure that out. We've been also working on improving that model into the system, so that when you're filing the issue, right as it notices that there might be a duplicate it will tell you, "there's this similar-looking issue here, if you want to look at it."

The interesting thing that comes from that is if there's going to be a lot of agents, or if you are farming out a lot of the work to these agents, then you might have like tens or hundreds of different conversations going with the agents as they're working on your issues. So then that, again, is a UI or UX problem; how do you manage that?

And then there's the second piece, which is these agent workflows. What we started seeing at the end of last year and early this year is that the models got better, and then we also saw the agents are getting better. We think maybe at the end of the year, we'll start seeing a lot more agents actually completing work, completing the issues, completing bugs or something else. 

Our strategy has to be more like, let's make Linear as the platform for those agents, and then also the platform for the humans to use them or manage them or delegate work to them. Because in the end, Linear is a platform for coordinating work. So you could coordinate it to a human, or you could coordinate it to an agent, or a human can coordinate their work, or delegate their work, to an agent.

What we've been working on is changing the API in such a way that it allows these agentic users, so that you can register [them] and they kind of have similar capabilities that normal users would have. But then the interesting thing that comes from that is if there's going to be a lot of agents, or if you are farming out a lot of the work to these agents, then you might have like tens or hundreds of different conversations going with the agents as they're working on your issues. So then that, again, is a UI or UX problem; how do you manage that?

Another interesting challenge is as a company, or if you're the owner or admin of this workspace, how do you then track or manage what these agents are doing? Or manage the scope of these agents, or even know which ones are installed? I think there's going to be a third category of problems that come from this that we need to solve to help companies manage them, budget them, maybe set certain rules or security controls, which is something we can help with. 

It feels like every company I talk to these days is trying to build a platform for managing agents. What makes you think that Linear can be the home for that kind of activity?

Our goal is to be this best-in-class tool for building products and planning. I don't think we'll be the platform for all kinds of agents, but we want to be the platform for building products. So when you think about any of these workflows of discovery, planning, building, and tracking, we want agents to be deployed in those workflows and those kinds of instances.

Our customers are the customers that are the first ones to use something. They're more like the pioneers; they're not like most legacy enterprises out there, they're more like the growth companies that can make decisions fast.

We already have this structured system interface, which I think would make it easier for people to adopt these things. Everyone's talking about agents, but not that many people are talking about how exactly will you use them, right? There's just not a lot of structure currently in this system. 

I would also say our customer base and developers are actually quite powerful. Our customers are the customers that are the first ones to use something. They're more like the pioneers; they're not like most legacy enterprises out there, they're more like the growth companies that can make decisions fast. We actually have an impactful customer base who is also willing to adopt new tools fairly quickly if they want to.

Also, because most startups use us, they also often integrate with us first, because that's kind of natural to them. What we've seen on the developer side — we track how many apps get created on the platform — is a 4.6x jump from February to March alone.

I think AI and these agents are driving a lot of people trying to build something on the API, and I think the API is also more modern and well designed, so it's actually easy to use. The positioning we have is being this more modern tool that these developers and growth and early-stage companies love, and then having that kind of purpose, [being] built for a specific purpose.

It's not like you can do anything on this platform; it's just more like a sharper blade, in a way, versus being this blunt system that can do all kinds of things.