Coinbase’s bet on agent-first development
Coinbase is one of crypto’s defining companies, with an ambition that stretches from the exchange to Base App and the broader shape of life onchain. Now, its product team is taking on the challenge of rethinking how software gets made when agents can do more of the work. Much of that experiment unfolds in Linear, which gives the team a place to hold the context as work begins to move differently.
When cameras first came out they were positioned as a way to draw without your hand. Computers were an improved calculator. Canny marketing, but more than that, a way for people to understand new technology using old mental models.
A similar thing is happening with AI today, seen in terms like “Copilot” and AI “assistant”, inviting you to do what you’re already doing, but faster, better, so everyone races to send more emails and auto-complete code.
Occasionally, new tools and the pioneers who use them break out and define new ways of doing things. Coinbase’s product team is one of those pioneers, and much of their work happens on Linear.
The clearest sign of that came in January 2026, when Chintan Turakhia, who leads AI acceleration at Coinbase, asked his entire engineering organisation to delete their IDEs and write zero lines of code for two weeks. The code editor is where the old way of working lives, so he took it away.
The experiment grew out of an observation about how large companies work. Thousands of people build different, overlapping pieces of the same product, and staying aligned across all of it becomes work in its own right. Teams buy visibility with standups, syncs, and status reports, and buy answers in Slack, pulling a teammate out of whatever they were building, a hidden tax paid in focus.
All that coordination leaves knowledge scattered across people’s heads, Slack threads, and meeting notes, a mess that humans can piece together and agents can’t, at least not very well. They need a single source of truth, and the way most companies work today is a wall for them.
For Coinbase, that source of truth is Linear. Product requirements, designs, bug reports, shipping status, and team structure all live there, because Chintan told his team from day one that agents would depend on it.
The team started with Linear’s own agents, then built Forge, their own, which can be called from Slack, GitHub, or Linear. When someone flags a bug in a channel, a workflow turns the message into a Linear issue, labelled, sized, and assigned, and Forge drafts a fix, opens a PR, and pushes it back to Slack for review.
Questions get the same treatment, status and timelines answered by Linear, and the code-specific things by Forge, whether that’s debugging a service or working out why a funnel is losing users. Nobody needed training into any of it, because you still just ask a question in Slack, the only difference is who answers.
Chintan pulls everyone along with Speedruns, near-weekly sessions where power users demo a new workflow and everyone on the call then runs it live. He hosted the first from his phone in the back of an Uber, walking 120 engineers through agents writing tests and drafting PRs, and they put up 80 pull requests in a sitting. By mid-year CEO Brian Armstrong was attending, one session produced over 500 PRs in fifteen minutes, and they’ve now taken GitHub down four or five times, which Chintan admits with a hint of pride.
Engineers now wake to the pull requests agents drafted overnight, review them, spin off ten or fifteen more for the day, and go heads down on the complicated stuff themselves. Development at Coinbase runs through the night, agents working while the team sleeps, and Chintan tracks autonomous operation time, how long an agent can run without a human stepping in, and that number keeps climbing.
The workflows have changed since January and they’ll look different again in six months. At Coinbase, every conversation now starts with the same question. “Why can’t an agent do this, and how do we make it possible?”