Coinbase did what most AI-workplace stories avoid: it put numbers on the productivity claim. The paper's May 10 brief on Coinbase's cohort split treated the company as the first weak link in a broader restructuring group. The earnings transcript now gives the more useful receipt.

CFO Alesia Haas said the company expects $50 million to $60 million in second-quarter restructuring expenses tied to headcount reduction. She also said Coinbase is moving toward an AI-native company, with pull requests per engineer up almost 80 percent year over year and integration test coverage across core services up threefold in six months. [1]

Brian Armstrong added a boundary that matters: product managers and designers may use AI agents to draft code, but human engineers still review all code before production, with multiple review levels for sensitive systems. [1]

The claim arrives inside a weaker quarter, not a victory lap. The same transcript shows total revenue down 21 percent sequentially, a $394 million net loss, and a softer crypto market. That makes the AI numbers less like corporate theater and more like management's explanation for how the expense base is supposed to change. [1]

That is the gap. X argues over whether AI code is reckless. MSM mostly files it under earnings color. Coinbase has made the debate auditable: costs, pull requests, coverage, and review rules can now be checked against later outages or margin claims.

-- DAVID CHEN, Beijing