Put yourself in a loop with your coding agents(github.com)

Put yourself in a loop with your coding agents

https://github.com/AsteroidHunter/expediter

3 コメント

I am trying to work with more agents and do so productively. I noticed that there are a few things that slow me down when working with n > 2 agents:

1. Not being able to monitor if the agent is working, stuck on a request, or done with a spec and awaiting instructions 2. The friction of finding the terminal window where the agent is working (I have too many tabs and apps open at any given time) 3. Time lost waiting for an agent to respond (especially true when designing a spec, even if they take a minute or two to think, that is one extra agent interaction window that you missed)

So, I designed a simple tool which addresses all the issues I face. I don't know if it is the ideal UX for interacting with many agents, but it has helped me more productively and continuously engage with agents I orchestrate! It has also helped me manage more agents than I otherwise would.

I would appreciate it if people give it try and share what they think of it + bugs if any + feature suggestions!

rpreetha2 か月前

Does the system optimize more for throughput or determinism/ reproducibility? Curious how those tradeoffs are balanced.

AsteroidHunter2 か月前

It is not much of a system, just a dev tool of sorts that is meant to help you increase your throughput when working with coding agents. As the user, you definitely have to calibrate how many agents you can actually work with without YOLO-ing on code quality.