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 listCommon 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.