Agentes de IA en paralelo con orbes en la nube

One of the most cracked engineers I know believes that local development is (mostly) dead. He walked me through why + his setup for using cloud agents in parallel... The big fucking problem: You've got one codebase on your computer. Worktrees are supposed to let you run agents in parallel. On a real stack they don't (database contention, port conflicts, a dev server that only works in one tree). So people flip back to one agent at a time and become the bottleneck themselves. Or they run multiple agents on the same machine: one writes a change, another writes over it, and they're fighting. Then you have to poke around, find everything each agent did, and test it independently. That's too much mental clutter. Locally you're operating like a CPU. You get an instruction, you build it, you queue a backlog. The solution: Stop sharing the computer. Give every agent its own. @seejayhess' setup is @AmpCode "orbs": a full computer in the cloud that clones the repo, starts the dev server, and runs one agent with nothing else contending. A "portal" is a live URL into that running app so you can click around from any device before you trust it. Isolated like an actual team, each person on their own machine. The payoff: It feels like a GPU, not a CPU. A list of 10 tasks, all started, real progress, loop closed. Smaller, contained diffs hit main faster and create fewer conflicts, not more. Favorite prompt: "give me irrefutable evidence that this works." On his laptop he would never run something so heavy he can't use Slack and Chrome. In an orb he will (millions of simulated DB writes, full demo videos). Smaller diffs, no mixed concerns, higher confidence, ship more often. Mental clutter is gone. One thread, one set of changes, one computer. He always knows the state. That's what lets him do more in parallel, not less. The last things he still does locally: read production logs, set a sensitive secret. He is not spinning up the app on his laptop.