How I’m Optimising My Grok Bot Workflow

I’ve been rethinking how my Grok Bot setup spends its effort. Instead of having one bot do all the planning, thinking and clicking, I’m trying a lean coordinator that hands focused tasks to other models.

Routine work goes to DeepSeek; architecture, design and review go to Claude Opus. For browser tasks, I’ve asked for Agent Browser, with connectors or scripts where they’re a better fit than repeated clicking.

The tradeoff matters: every handoff adds context and another round trip. Delegating everything isn’t automatically cheaper or better.

So I’m adding attribution: log each model call under the bot that requested it, and report which model actually ran—not just which one was planned. I’m also keeping usage from different tools separate rather than hiding it in one total.

No savings claim yet. The instructions are written; the next test is whether the logs show they’re being followed.