Supercharge your coding agents: DESIGN.md, terminal diffs & Claude tips

Google's DESIGN.md

We are currently testing Google's DESIGN.md specification to see how it performs and scales within real-world development projects. Stay tuned as we analyze its impact on structuring software architecture documentation for AI agents and human teams alike.

Terminal diff reviewer by Todoist's CEO

Amir from Todoist has built a slick terminal diff reviewer. The workflow is straightforward: you mark code issues inline directly within your terminal layout, and then let a dedicated coding agent step in to fix them automatically.

The hidden cost of not using English in LLMs

Aran Komatsuzaki shared a fascinating breakdown regarding non-English token inefficiency. While Chinese models run cheaper natively, Gemini and Qwen impose the lowest "non-English tax" overall. On the flip side, Anthropic features the highest tax by far, with Hindi ranking as the least optimized language.

Claude workflow tips from the team

• Carlos López: Use ! to execute terminal commands directly within context. Running ! npm run test and letting Claude fix those specific errors saves massive time and token consumption compared to asking it to broadly review unit tests.

• Maxime Ancelin: Resume your last session quickly via CLI using claude -c.

• Juan Artero: Use /rename [name] to rename sessions. Matching the session name to your active git branch helps isolate independent tasks perfectly.

Bonus: Anthropic just bumped Claude Code weekly limits by a massive 50%!