> I didn't want to be a project manager for agents

> Running a bunch of agents feels less like being a project manager for agents and more like stepping into this super mech suit where suddenly I don’t just have two arms, I have twelve.

But then recently I asked Claude Code to upgrade a particular library for me that I had forked and to reapply the changes I had made in my fork. It just did it as if by magic. As if it already knew everything about the original library. Given that the library is open source and on Github then Claude probably does know everything about it.

But then I can still have times when Claude Code does something really helpful (e.g. at work it helped create the first version of an MCP server for our own agentic application), while other times it can do some downright dumb things (like editing directly a library's code so something worked rather than changing my code so it worked).

Any time I'm feeling like the machines are about to take over I just do this: 1. Ask Claude Code to write some tests. 2. Same model, but different window, ask it to delete any tests that were just added but which don't test anything substantive.

It always finds something to delete that it had just added.

I'm not even talking about using a different model for code review or anything like that. Just asking the exact same model 10 seconds after it wrote some tests to delete the bad ones.

So I'm not yet at the slightly panicked stage :D