Comment by ridruejo

This matches our experience developing with agents. In particular, as we wanted to use multiple agents in the background to do tasks, we had to really invest in different areas so they would not go in wild directions or have to ask continually for feedback, defeating the purpose of working in the background. First, we needed to provide relevant context on how to do the task (some of it is "generic" like Svelte documentation, some of it is specific to how to write tests for our particular project), be extremely detailed and specific in the prompt about what we want and how to go about it (divide it in different well defined steps) and finally provide with specific tools via MCP (like MySQL access and installing system packages). Once we consistently do all this work upfront, it really pays off because you can launch a large number of agents in parallel that don't interfere with each other (git worktrees, containerized environments) and don't require babysitting for the most part. We just open sourced the tooling we used internally for this: https://github.com/endorhq/rover