Route each task to the cheapest capable model

A user changed OpenClaw so it could choose lower-cost models for CRM work and email drafting, with manual approval before any outside action.

Done with OpenClawยท Advanced

What they did
Koka Sexton used OpenClaw as a routing layer across OpenAI, Claude, and Gemini. He then had the agent alter its own code to add a manual throttle that selects the least expensive model that can still handle the job. In practice, it compares tasks like a 30-row CRM review and a batch of 20 outbound emails, and it can switch models when he asks it to.
How it went
The setup now ties model choice to task difficulty, while OpenClaw also handles follow-ups and draft writing like a chief of staff. The source does not give any savings numbers. External messages still wait for his explicit approval.
Tips
Keep the switching rule manual if you want a human in the loop for model changes. For anything that sends, submits, deletes, or pays, have the agent pause and ask before continuing.

Try it yourself with OpenClaw

You are my model router inside [system name]. Review the task type and pick the cheapest model that can handle it well from this list: [models and prices]. First, show your plan, the code changes you would make, and a short test on sample tasks such as [task A], [task B], and [task C]. Pause and wait for my approval before you apply any change that could send, submit, delete, pay, or message anyone outside the system. After approval, update yourself, run the real tasks, and report which model handled each one and the estimated cost.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on substack.com โ†—

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.