What are the main AI orchestration patterns?

Microsoft’s catalog of agent orchestration patterns lines up with what most teams converge on.^[1]

• Sequential (prompt chaining): one step feeds the next, often with a gate between them.
• Parallel (fan-out / fan-in): independent sub-tasks run at once and the results are merged. Useful for sectioning a problem or for voting across several attempts.
• Handoff / routing: a triage step sends the work to the right specialist.
• Hierarchical (orchestrator-workers): a coordinator breaks a goal into pieces, hands them to worker agents, and assembles the result.
• Evaluator-optimizer: one step generates, another checks against criteria, and the loop repeats until it passes or hits a cap.