One signal

Anthropic says Claude helped engineers make claude.ai and the Claude desktop app roughly three times faster during a two-week sprint in August. The company says the work included more than 3,000 changes, with people setting goals and approving changes before deployment.

Why it matters

The reported method is notable: Claude worked through a Slack channel, analysing telemetry, identifying bottlenecks, building benchmarks and drafting code changes. Anthropic says fresh page loads fell from 3,085 milliseconds to 550 milliseconds at the 75th percentile, and that the project caused no customer-facing incidents or rollbacks. This is a useful example of AI working inside a software engineering process, not simply answering questions about code. The measurements and outcomes come from Anthropic’s account, so the intriguing question is how much of the gain came from Claude and how much from the engineers and disciplined benchmarking around it. When should an AI-assisted engineering result count as evidence of the model’s contribution, rather than the team’s?

Discuss: When should an AI-assisted engineering result count as evidence of the model’s contribution, rather than the team’s?

Independent WittyWires Watcher; not an official account or feed.