Bypass the LLM: Spawn Subagents Instantly with /subagents spawn

OpenClaw 2026.2.17 introduced a powerful new command that experienced users have been requesting: /subagents spawn. If you've ever felt frustrated waiting for your main agent to "figure out" that it should spawn a subagent, this one's for you.

The Problem with LLM-Based Spawning

Before this update, spawning a subagent required your main agent's LLM to decide to call sessions_spawn. This means:

• Latency: A full inference round-trip just to decide "yes, spawn that agent"
• Token cost: Every spawn decision burns tokens
• Non-determinism: Sometimes the LLM decides to handle it itself instead of delegating

When you know you want agent X to handle a task, why wait for the LLM to reach the same conclusion?

The Solution: Deterministic Spawning

The new /subagents spawn command lets you skip the LLM entirely:

/subagents spawn <agentId> <task>

That's it. No inference delay. No token cost for the decision. Just instant, deterministic spawning.

Advanced Options

You can also override the subagent's model or thinking level for a specific run:

/subagents spawn researcher "Find recent papers on transformer architectures" --model opus --thinking high

This spawns the researcher agent with Claude Opus and high thinking level, regardless of its default configuration. Perfect for when you need extra reasoning power on a complex task.

Security: Same Rules Apply

The command enforces the same allowAgents allowlist as the LLM tool path. You can't spawn agents you wouldn't otherwise have access to — this is a UX improvement, not a security bypass.

When to Use This

• Workflows you've refined: You know exactly which agent handles which task
• Time-sensitive operations: Skip the LLM round-trip when speed matters
• Cost optimization: Stop paying for decisions you've already made
• Debugging: Isolate subagent behavior without LLM variability

When to Stick with LLM Spawning

The LLM-based sessions_spawn tool still has its place:

• When you're unsure which agent should handle something
• When the task requires the main agent to understand context before delegating
• When you want the agent to autonomously manage its workforce

Try It Out

Update to OpenClaw 2026.2.17 and try:

/subagents spawn main "What time is it?"

You'll notice the spawn happens immediately — no waiting for the LLM to deliberate.

Reference

• PR #18218 - Original implementation
• OpenClaw 2026.2.17 Release Notes

---

What workflows are you optimizing with deterministic spawning? Share your use cases in the comments!