The Claude for Open Source program is opening up to a wider range of contributors, offering 6 months of Claude Max 20x free to maintainers, core contributors, and anyone keeping critical packages running.
The Claude for Open Source application process has been updated to be faster and more consistent. Maintainers who were previously rejected are encouraged to reapply.
If a maintainer was accepted to Claude for Open Source but let their code expire before using it, the fix is simple: submit a new application.
Anthropic published the first public telling of how Claude Code was built, tracing its origins back to Anthropic safety research. bcherny framed it as only the beginning, writing that the team is "1% done."
Before writing any code, asking Claude Code to produce a low-fidelity artifact showing how the UI will look and how systems fit together moves iteration from the code stage to the planning stage. The key instruction is to keep the plan and artifact in sync as the conversation evolves.
A developer built a tool with a drag slider that transitions the view between old and new code states. The motivation: iterating on features in already-written code with Claude gets expensive, so moving iteration earlier in the process saves tokens and money.
Gradually visualizing diffs as they're applied is being explored as a way to make code changes more legible. The original builder spent only a few minutes on it and is eager to see results from others trying it.
Anthropic published a retrospective on how Claude Code was built, told by early team members. Cat Wu shared it with a brief note calling it a look back from the early team.
Bcherny confirmed he has never used Aider, but noted that some people who worked on Clide, the internal predecessor to Claude Code, had used it. The lineage runs through Anthropic's own internal tooling rather than the open-source agentic coding scene.
Claude Design can produce animations and video when prompted directly, which is how the animated UI mockup GIFs in the thread were created.
