Anthropic uses Claude Code for 90% of coding and code review

11 articles · Updated · O'Reilly Media · Apr 29

Product lead Cat Wu said engineers now produce about 200% more code than a year ago, with five to 10 review agents running in parallel.
The company says AI review traces changes across files, catches adjacent bugs before merge and supports a model where PR authors retain end-to-end ownership, including post-deployment issues.
Anthropic says internal uptake began as an API test by Boris Cherny, spread company-wide through rapid feedback, and helped justify an external release while shifting engineers toward judgment and product taste.

As AI writes 90% of code, how do we prevent a future where engineers no longer understand the systems they are building?

If 'product taste' is the new top engineering skill, how can it be systematically taught and measured across an entire organization?

After AI agents erased $830 billion from software stocks, which white-collar industry is next to face a similar disruption?