Multi-agent orchestration

Instead of one assistant doing everything in one context, several agents take roles: one plans, others research or write in parallel, one reviews. An orchestrator hands work out and collects it. It is how long or wide tasks get done without the single context filling up, at the cost of coordination.

Standing: Rising since 2025-06
Stars, week: +5,592 across 5 repositories
Stars, month: +28,261, through 2026-11-08
Listings: 6 chosen by hand

The standing

Where it stands

Among the 14 concepts. Bars: GitHub stars gained this week.

Established · 7

Rising · 7

How to use it today

In ChatGPT or Claude, as a person doing the work, not as a developer.

1In Claude Code, use subagents (or ask "spin up a subagent to research X while you draft Y") when a task has independent parts.
2Give each agent one role with its own instructions and its own output format; the orchestrator should merge formats, not personalities.
3Keep the number small: two or three roles cover most work; more mostly adds handoff errors.
4Have one agent whose only job is to review the others' output against the original brief.
5Read the merged result as a whole before using it; seams between agents are where contradictions hide.

What changes day to day

You start writing briefs for roles rather than instructions for steps. Large pieces of work (a market scan across twenty companies, a spec with five sections) finish in one sitting, and your work shifts to the brief and the final read.

Where it goes wrong

Agents that share no context repeat each other's research or contradict each other.

Orchestration multiplies cost and time when the task was sequential anyway.

Listings that use it well

Chosen by hand after reading them, each with the reason. Install from the listing's own page.

Listing	Kind	Why it is here	Stars
superpowers	Agent skills	dispatching-parallel-agents and subagent-driven-development split a plan into independent tasks executed by subagents.	315k
ai-agent-book	Agent skills	multi-agent-design covers peer, manager and decentralised topologies, shared context decisions and failure modes.	57k
engineering	Agent skills	agent-designer picks supervisor, swarm or pipeline patterns; agent-workflow-designer writes handoff contracts; agenthub races parallel subagents in worktrees.	29k
c-level-advisor	Agent skills	A virtual board: chief-of-staff routes to advisor roles and board-meeting runs a six phase deliberation across isolated C-suite agents.	29k
agent-teams	Agent skills	Team composition, task coordination and communication protocols for Claude Code agent teams.	42k
oh-my-claudecode	Agent skills	Teams first orchestration with team, swarm, pipeline and autopilot modes.	41k

Elsewhere

Repositories that embody the idea and are not in the catalog.

crewAIInc/crewAI ↗Role-based multi-agent framework with crews and flows.
microsoft/autogen ↗Microsoft's multi-agent conversation framework; in maintenance since its merge into Microsoft Agent Framework in October 2025.

Standing

Our judgment, with the reason and the sources it rests on.

Rising since 2025-06. Anthropic described its orchestrator and subagent research system in June 2025 and shipped custom subagents in Claude Code in July 2025. Cognition argued against multi-agent systems the day before; its April 2026 follow-up found that many agents can research and review in parallel as long as only one of them writes. That is now the common advice. Adopted faster each month; the vendors are building for it and new tools keep appearing.

How we built our multi-agent research system2025-06-13
Don't Build Multi-Agents2025-06-12
Enabling Claude Code to work more autonomously2025-09-29
Multi-Agents: What's Actually Working2026-04-22