3 Comments
User's avatar
Suhrab Khan's avatar

This approach is brilliant! Turning coding agents loose asynchronously transforms curiosity into real experiments. The human role shifts from execution to insight, making exploration far more scalable and efficient.

JP's avatar

The fire-and-forget pattern you describe for coding agents maps surprisingly well to how hooks work in Claude Code now too. Since this was published they've shipped async hooks that let logging, metrics, and notifications run without blocking execution. Same principle, different layer. Wrote up the specifics of when each hook event should go async vs sync: https://blog.devgenius.io/claude-code-async-hooks-what-they-are-and-when-to-use-them-61b21cd71aad?sk=40d0b1c86232a799e19b20ba53b8e2c3

Alex Casalboni's avatar

I’ve been doing this for a few days. It feels like a superpower 🚀💪 I struggled a bit initially with Codex’s network restrictions and some weird caching issues when using private repositories, but now it works great! Firing up ideas while on mobile and following up later directly on GitHub is perfect for research projects and quick experiments ⚡️