Agent Orchestration
How to run more than one coding/ops agent: one “god” router vs a swarm of specialists, parent/child sessions, permissions, evals, and token budgets.
Definition
AI-Assisted Development is driving a single coding agent well. Agent orchestration is the layer above that: routing work across agents, isolating data, scheduling loops, and deciding whether the user talks to one ambient assistant (Guillermo Rauch / Vercel: one “god agent”) or a team of named specialists (Riley Brown’s operator instinct; Greg Isenberg’s parent/child sessions and playbooks). Still emerging — three adjacent sources, no official standard.
Key Properties
- Design fork: one router vs named swarm.
- Recurring machinery: skills/tools/permissions, evals, token budgets, scheduled loops.
- Adjacent to CI/CD & Automation (DORA Elite) when agents triage PRs or watch production.
In the Sources
| Source | Context |
|---|---|
| The AI Agent Every Company is About to Build | Vercel CEO Guillermo Rauch | Rauch: one ambient UX that routes. |
| Most Valuable Skill of 2026: Managing AI Agents | Playbooks, parent/child, watchdog. |
| Why Your AI Offer Isn't Selling, and How to Fix That | Auto-router, small AI-native teams. |
Related
Concepts: AI-Assisted Development, CI/CD & Automation (DORA Elite) Entities: Vercel, Guillermo Rauch, Claude Code, Greg Isenberg