Agents in action
Anthropic let Claude speed up claude.ai: 3,000+ changes, page load from 3.1 s to 0.55 s
In an August sprint, Anthropic's team ran Claude in one Slack channel with standing instructions to make claude.ai and the Claude apps faster. Humans set the goals and approved every change.
A before/after video from the same sprint, posted by @ClaudeDevs: long answers stream about 4× smoother.
What the agent did
- Used Datadog data to pick the 4 highest-impact user journeys and turned them into 13 measurements
- Hit 12 of the 13 targets by day three
- Ran more than 150 threads at once at peak and merged 3,000+ changes, over 200 on the busiest days
- Shipped it all with no customer-facing incident or rollback
Why it matters
Fresh load of claude.ai went from 3.1 s to 0.55 s, sending a message in a Cowork cloud session from 928 ms to 48 ms, and the 13 measurements improved about 3.1× overall. Once a goal is a number, agents can grind it down in parallel while people review.
The model was an internal research model the team calls roughly comparable to Opus 5.5. Every PR had automated review plus at least one human approval.
Source: Claude engineering blog
Every number on this page comes from the source. Our charts redraw those numbers; videos are embedded from the publisher.
Agents in this story
| # | Agent | Gid Score | Downloads 7d | 7d | 30d | Stars | Latest release | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| 3 |
Claude CodeAnthropic |
80 | 11.7M | down 20.3% | down 29.7% | 150K+170/day | 2.1.295yesterday | ||||
| 187 |
Claude CoworkAnthropic |
20 | — | — | — | — | — | ||||
| No agents match that filter. | |||||||||||
