How Coinbase Scaled AI to 1,000+ Engineers

How Coinbase Made AI Stick Across 1,000 Engineers

Most large engineering organizations try AI tools, see a brief uptick, then watch adoption flatline. Chintan Turakhia, Senior Director of Engineering at Coinbase, went through that exact trough of sorrow — and came out the other side with a playbook that actually worked. In this conversation with Claire Bell on How I AI, he walks through the specific tactics that drove real adoption.

On why adoption doesn’t stick: “The company tried to adopt other AI tools and we saw this uptick in adoption. People opened it up, checked the box, did kind of like a hello world thing, but it didn’t stick. My biggest thing is, how do I make this damn thing stick?” The problem wasn’t the tools — it was the models weren’t ready in late 2024, and once one engineer bounced off, the entire team wrote it off. Turakhia’s mental model: the foundational models will always get better, so build the reps now.

On leaders writing code again: Turakhia spent January through April 2025 in Cursor every single day, personally fixing bugs, putting up PRs, and showing engineers what was possible. The worst thing a leader can do is decree AI adoption from a meeting room. “I do think that it’s really important when you’re doing this organizational transformation that you have a single person with incredible conviction at the leadership level who is also hands on the metal.” He started with toil — unit tests, linting, git commands — the soul-crushing work no one wants to do.

On PR speedruns: The breakout moment was a timed event: every engineer picks up a trivial bug or copy change and puts up a PR using Cursor. In 15 minutes, 100 engineers submitted 70 PRs. They broke GitHub’s infrastructure. Then they went company-wide: 800 engineers, 400 PRs in 30 minutes. “It was really sort of a death to status updates, long live building moment.”

On measuring what matters: Turakhia obsesses over one metric: time from ticket to when the change lands to the user. That encompasses prioritization, coding, review, and deployment. PR review cycle time dropped from 150 hours average to roughly 15 hours — a 10x improvement. The goal: someone gives feedback, and the fix ships before the call ends.

On Cloudbot — their internal agent: Coinbase built an in-house Slack bot that chains Linear tickets, MCPs (Datadog, Sentry, Amplitude, Snowflake), and multiple codebases. The workflow: capture user feedback via audio → LLM extracts bugs → creates Linear ticket → agent generates PR → deep links to Cursor branch → QR code for mobile testing. One night, Turakhia kicked off 200 bug fixes from the feedback tool in a single batch.

On the “super builder” role: Turakhia invented a new role: the super builder. Its single most important job is to create more super builders. These are the people who drive AI tooling, build internal agents, and accelerate everyone else. His advice: being among the top three most AI-proficient people in your engineering org is one of the best career moves you can make right now.

On coordination overhead vanishing: “My calendar is empty. Almost empty. And the reason why is because the coordination overhead of prioritizing, changing the roadmap — no, you just do things.” Leaders are writing more code. Teams skip the sprint planning debates because feedback-to-fix cycles are measured in minutes, not sprints.

6 Takeaways from Coinbase’s AI Adoption Playbook

• One conviction-driven leader on the metal — Adoption requires a hands-on champion who codes daily, not a mandate from above
• Start with toil — Unit tests, linting, git commands — remove the soul-crushing work first, and engineers will lean in
• PR speedruns create breakthroughs — Timed events where everyone ships at once build conviction and expose infrastructure limits
• 10x faster PR reviews — Cycle time dropped from 150 hours to 15 hours by compressing the entire feedback-to-ship pipeline
• Internal agents beat external tools — Cloudbot chains Linear, Slack, and MCPs to go from user feedback to merged PR autonomously
• “Super builder” as a career path — The person who makes everyone else more productive with AI is the most valuable hire right now

What This Means for Engineering Organizations Adopting AI

Coinbase’s story is significant because it’s a real case study at scale — not a 5-person startup, but 1,000+ engineers at a public company with serious security and compliance requirements. The playbook is replicable: start with a conviction-driven leader who codes, target toil first, create visible wins through speedruns, build internal agents where external tools can’t reach, and measure the one metric that matters — time from feedback to user. The organizations that figure this out don’t just ship faster. They fundamentally change what’s possible with their existing headcount.