Anthropic says 80% of its new production code is authored by Claude
The number is a claim about a single company with unusual access, unusual incentives and unusual tooling. It is still the most concrete public figure anyone has put on agentic coding in production.
Take the caveats first, because they are load-bearing. This is the model's maker, reporting on itself, with no external verification, and authored is not the same as designed, reviewed or shipped. A figure produced under those conditions is evidence about what is possible in the best case, not a benchmark anyone else should expect to hit.
With that said, 80 percent is a large number even heavily discounted, and it comes from an organisation that would be embarrassed by it if the code were bad. The interesting question is not whether other companies will reach it but what changes in the surrounding practice — review, testing, architecture ownership — when most new code arrives already written.
The honest gap is that nobody has published the other half: defect rates, review burden, or how much of that 80 percent is boilerplate that a generator would have produced in 2015.
VentureBeat — Anthropic says 80% of its new production code is now authored by Claude → · Anthropic — 2026 agentic coding trends report →