ai-agent-orchestration

name	ai-agent-orchestration
description	Coordinate multiple AI agents or subagents on software work without duplicating effort or causing conflicts. Use when Codex is explicitly asked to delegate exploration, implementation, review, testing, integration, or parallel analysis across a codebase.

AI Agent Orchestration

Purpose

Use additional agents to increase throughput and independence, not to outsource the critical path blindly. Bounded worker ownership is a coordination tool, not a claim that the solution must be small; broad architecture pivots can be split into disjoint ownership areas under one explicit design.

Orchestration Workflow

1. Decompose the task.

   • Identify the immediate blocker that the lead agent should handle locally.
   • Identify sidecar tasks that can run independently.
   • Identify write scopes that can be made disjoint.
   • If evidence calls for a subsystem replacement, decompose by architecture boundary rather than shrinking the goal into unrelated local tweaks.

2. Assign ownership.

   • Give each worker a bounded module, file set, or question.
   • Tell workers they are not alone in the codebase.
   • Tell workers not to revert edits made by others.
   • Require workers to report files changed, tests run, and remaining risk.

3. Keep exploration concrete.

   • Good: “Find every path that imports _stageView_pixar_orig.py and report whether it is safe to delete.”
   • Bad: “Understand the whole repo.”
   • Good: “Patch only nanousdview/python/.../usdviewq to remove Hydra debugger artifacts.”
   • Bad: “Clean up legacy code wherever you see it.”

4. Avoid duplicate work.

   • Do not ask several agents the same unresolved question unless independent confirmation is the goal.