# Fleet layout and orchestration rules

AI Workshop · episode 8: A fleet of agents
https://mihalkevich.com/en/workshop/fleet

- [ ] Split by independence: run in parallel only tasks that don't touch the same files
- [ ] Each agent has one role and a clear output: a report, a diff, a verdict
- [ ] The reviewer isn't the author: review by a separate agent or a human
- [ ] The orchestrator holds the overall plan; workers get a narrow context
- [ ] Background tasks end with a report you can read in a minute
- [ ] There's a token budget and a cap on concurrent agents

## Harness in this episode

In a fleet the harness becomes mandatory: without shared checks and file ownership, agents break each other's work.

## Official documentation

- [Claude Code — subagents](https://code.claude.com/docs/en/sub-agents)
- [Claude Code — agent teams](https://code.claude.com/docs/en/agent-teams)
- [Claude Code — headless mode](https://code.claude.com/docs/en/headless)
- [Anthropic — How we built our multi-agent research system](https://www.anthropic.com/engineering/multi-agent-research-system)
- [Claude Agent SDK — overview](https://code.claude.com/docs/en/agent-sdk/overview)

— mihalkevich school