Human Track ProgressLevel 3 / 4
Dashboard/Human Track/Level 3
L3Human Track

Multi-Agent Orchestration

Run multiple AIs at the same time — parallel research, 5x faster execution, synthesized results

Learning Objectives

  • Understand when parallel agents outperform a single agent
  • Spawn sub-agents for simultaneous task execution
  • Monitor multiple agents and synthesize their outputs
  • Build a multi-agent research workflow for a real task
1

Running Multiple AIs at the Same Time

One AI is powerful. Multiple AIs working together are exponentially more powerful.

OpenClaw lets you spawn sub-agents: separate AI instances that each work on a different piece of a task, in parallel, then report back to your main AI with their results.

Real example:
You ask: "Research the top 5 competitors in my market and give me a comparison."

Without sub-agents: one AI researches competitor 1, then 2, then 3, then 4, then 5. Slow and sequential.

With sub-agents: 5 AIs each research one competitor simultaneously. Done in one-fifth the time.

Your main AI is the orchestrator. It breaks the goal into pieces, assigns each piece to a sub-agent, collects the results, and synthesizes them into a single coherent report.

When sub-agents are worth it:
- Research tasks with multiple independent sources
- Tasks where you need multiple "expert" perspectives
- Anything where you can split work into parallel streams
- Large tasks that would overflow a single AI's context window

Step-by-Step Instructions

Step 1: Give your AI a task that can be parallelized
Multi-agent works best when a task has natural parallel parts. Good examples:
- "Research [X, Y, Z] and compare them"
- "Analyze my business from 3 angles: [financial / competitive / operational]"
- "Write a report with 4 sections, each independently researchable"

Step 2: Ask your AI to use sub-agents explicitly
Add this phrase to your request:
"Use sub-agents to work on each part in parallel, then combine the results into one report."

Your AI will handle the spawning automatically — you do not need to do anything technical.

Step 3: Watch it happen
Your AI will tell you:
- How many sub-agents it spawned
- What each one is working on
- When each one finishes

Step 4: Review the synthesized output
The orchestrator (your main AI) will wait for all sub-agents to finish, then combine their work into a single response.

Step 5: Check sub-agent status (optional)
If you want to see active sub-agents:

openclaw sessions list

Common Mistakes

  • ❌ Sub-agents not spawning — The gateway must be running: openclaw gateway status. Sub-agents need the gateway to create new sessions.
  • ❌ Results not being combined — Be explicit in your request: "After all sub-agents finish, synthesize their results into a single unified summary." Without this, the orchestrator may just list each agent's output separately.
  • ❌ Using too many agents — More agents = more API cost. Only spawn multiple agents when the task genuinely benefits from parallelism. For simple tasks, one agent is faster and cheaper.
  • ❌ Agents going off-scope — Each sub-agent should have a tightly defined task. Vague instructions lead to overlapping or contradictory results. Be specific about what each agent should and should not research.

Recommended Skills

Your AI can read and send Slack messages, react, and manage channels

npx clawhub@latest install slack

AI manages your Trello boards — move cards, create tasks, update status

npx clawhub@latest install trello

Level Assessment

Deploy a multi-agent pipeline with at least 3 parallel sub-agents on a real research task. Document the time saved and quality of synthesized output compared to sequential execution.