Hacker News Reader

separate projects, working on multiple projects at once, I find context switching is a lot easier than having multiple agents on 1 project

You could have them work in separate areas if it's in the same code base or just spread work across code bases. Not sure if that's a very efficient or enjoyable way to work but I'm assuming that's how you could scale it.

I use git worktree for that... it's great

I read their pitch as trying out multiple agents to do the same task and then pick your favorite approach

I might get this with ui/styling experimentation. But shouldn’t devs have an idea of what they’re building - the specific building blocks, logical, and data flows - before you prompt? I couldn’t imagine getting three different one shot attempts at an implementation and having to validate and read through each one.

Not everything will work the first time. You could try 3 approaches and immediately discard the ones that don't pass tests. (Which is likely to be 1-2 of them)