Lesson 3 / 7 · ~12 minutes
Cloud: delegate
and let it work
This is why Codex exists. A cloud task runs in an isolated environment on OpenAI's servers with a copy of your project — you can switch your own computer off meanwhile. And since every task gets its own sandbox, you can happily send three at once.
Local or cloud?
- Local — small things you want right now: a typo fix, a tiny tweak, a quick question about the code. You see it immediately.
- Cloud — anything that takes time: a new feature, tests for a whole module, refactoring, bigger changes. Brief it, walk away, come back to a finished result.
What you get back
A finished task returns three things: the diff (the exact list of changes — green added, red removed), the logs (what the agent did and why) and the test results. Your job is the review: go through the diff, verify the behaviour, decide. If you don't like the result, you fork the task — adjust the brief and the agent tries again, without losing the original version.
Parallel delegation, live
The replay: three tasks for the rate calculator, briefed at once — watch them come back one by one and see what diff review looks like.
Decision game: what do you do?
Three situations from a delegator's life. Pick the best reaction.
1. Which task is worth sending to the cloud?
B — long, self-contained work with a clear end is ideal for delegation. A typo you fix locally in seconds; sending it to the cloud is like calling the movers for one chair.
2. You have three unrelated tasks. How do you brief them?
B — unrelated things = separate tasks. They run at once (faster), each gets a clean diff (easier review), and if one fails, the others aren't held up.
3. The task finished, tests green — but you don't quite understand the diff. Now what?
C — that's exactly what forking is for. The agent will explain the changes in plain language or propose a simpler approach. Blind accepting and blind discarding both rob you of control.
Summary
- Small things locally, anything longer to the cloud.
- Brief unrelated tasks separately — they'll run in parallel.
- Review = diff + logs + tests. Don't understand? Fork and ask.
- Mobile: briefing, watching, and Remote control of a paired computer.