Why run things in parallel?
Hand off small, well-scoped jobs to cloud agents.
Cloud agents run tasks in their own sandboxes, in parallel, in the background. You submit, step away, come back to diffs / pull requests.
| Product | From | What it's good at |
|---|---|---|
| Devin | Cognition | Autonomous engineer over a whole repo |
| Claude Cowork | Anthropic | Many small tasks across one or more repos |
| Copilot Workspace | GitHub | Issue → PR with task plan |
| Codex Cloud | OpenAI | Background coding with PR review |
| OpenHands (cloud) | All Hands AI | Open-source alternative, self-hostable |
Coding CLI
Pair programming. You and the agent on the same thing, together, in real time.
→
Cloud agent
Delegation. Many tasks in parallel sandboxes. You review when convenient.
Best for
- Small features, bug fixes, dependency bumps
- Refactors with a clear before/after
- Documentation passes, test additions
Not for
- "Design this product" — ambiguous, open-ended work
- Architectural debates that need conversation