I’ve tried Conductor and other git worktree based approaches. Multiple checkouts are easy, but you’re still constrained by the same local dev setup (DB, services, ports). Spinning up multiple local full stacks also hits resource limits pretty quickly.

Cloud based agent setups feel too black-boxy. You often need a real environment to iterate, even for the agent: debug, inspect logs, run the dev server, and use the frontend to actually verify behavior. Without that shared runtime, the agent’s feedback loop isn't tight enough, and you can't verify changes as well.

Feels like the bottleneck is isolated environments that can run concurrently, but with tooling to make it more like multiple local setups instead of black-boxed agent environments.

Has anyone found a practical way to run multiple agents in parallel?