Coinbase CEO Brian Armstrong wants internal teams to compete like startups, with AI agents as the referees

Brian Armstrong has a management philosophy that sounds a lot like natural selection for corporate infrastructure. The Coinbase CEO is pushing for the company’s internal platforms to compete against each other the same way external products fight for customers, with AI agents serving as the mechanism that evaluates performance, routes work to the best option, and sidesteps anything that isn’t pulling its weight.

The AI workforce inside Coinbase

Coinbase currently operates roughly 1,200 full-time AI agents integrated across its internal communications and workflows. These aren’t simple chatbots answering FAQ questions. They handle code reviews, provide design input, evaluate strategy, and participate in team collaboration in ways that mirror what a human colleague would do.

The results have been tangible. Coinbase has reported approximately a 2x year-over-year increase in the amount of code shipped per developer, alongside a decrease in bugs reported.

Armstrong has gone further, predicting that AI agents at Coinbase could eventually outnumber human employees. It reflects a deliberate restructuring strategy aimed at making the company leaner and faster through deep AI integration.

Earlier this year, Coinbase began testing AI agents modeled after former executives. The concept is unusual enough to warrant a double take: digital replicas of past leaders contributing to strategic discussions and collaboration.

Internal competition as an operating principle

The core of Armstrong’s advocacy goes beyond simply deploying AI. He’s arguing for a fundamental shift in how large companies think about their own internal services. By letting internal platforms compete, with AI agents capable of comparing services and routing work to the strongest option, underperforming tools face actual consequences. They lose usage. They get bypassed. Eventually, they get replaced.

What this means for Coinbase and the broader industry

For Coinbase specifically, a company that can ship twice as much code per developer with fewer bugs is a company that can iterate on products faster, respond to regulatory changes more quickly, and roll out new features ahead of competitors.

The risk, of course, is that aggressive AI integration introduces new failure modes. AI agents making routing decisions about internal infrastructure could create blind spots that human managers would catch. Modeling digital agents after former executives raises questions about institutional bias being encoded into decision-making systems. And the prospect of AI agents outnumbering human employees, while efficient on paper, creates organizational dynamics that no company has fully navigated yet.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.