An AI bot as a colleague

If you haven't come across OpenClaw yet, it's the open-source AI agent that went viral earlier this year. It runs locally or on virtual servers and can actually do things on your machine, not just chat about them. ComfyUI is the node-based tool that a lot of the serious AI image and video work runs through these days.

I had avoided learning ComfyUI for two years, but then I got on an OpenClaw bot to set it up for me, and it all felt strangely like working with a real colleague...

For years, I knew Comfy was powerful, and I could see people making tons of incredible stuff with it. But everyone said the learning curve was steep, and I already had too much on my plate to start weeding through Youtube tutorials.

When the OpenClaw craze started in January, I was not very impressed by the stuff I saw (bots handling email, shopping groceries etc), as it all felt like trying to find a use case for a technology. But I kept thinking about what COULD be helpful to me, and then it hit me: what if I didn't use it as an errand boy, but as a colleague?

Specifically, a Technical Director.

In animation, the TD builds and maintains your pipeline (the technical backbone that lets artists actually make things). At a studio, you go over to the TD, explain what you need, they go build it, and you come back to review. It's a collaborative loop of briefing, building, reviewing, iterating.

So I set up an OpenClaw bot and gave it a persona, job description and named it Frank_TD. I told Frank about the animation project I'm working on and asked it to help me test a ComfyUI pipeline for it. Comfy now has a Cloud API, so I plugged that in and gave Frank access. Then we just... worked together, through natural language in Telegram, like I might chat with a real colleague. We specced out the hypothesis for what we wanted to test, and Frank researched SOTA approaches on its own. And then I said: okay, go build it. Report back when you're done.

And it did. It built the pipeline and uploaded the first generated image right into our Telegram chat.

I'll confess that there was some virtual high-fiving.

After that, I loaded up what Frank had built in ComfyUI Cloud and started tweaking things to my liking. Did I learn how to set up ComfyUI from scratch this way? No. But I've spent 25 years in animation studios where you split your tasks. Our process felt like being a producer, reviewing a TD's setup and saying "okay, change this, try that, let's go."

The whole thing cost me maybe $30/mo for the virtual server droplet and a few dollars in API costs.

That number matters here, since I'm at a stage with this project where I don't have funding to hire outside help. Normally, that means certain things simply don't get done: you can't test pipelines that you can't build, and you can't build pipelines you don't have time to learn. But $30 and a bit of back-and-forth with Frank got me to a working setup I could evaluate and iterate on.

This is what people miss when they talk about AI agents shopping groceries: The real value is in replicating the division of labor that makes real productions work. A producer doesn't need to know how to rig a character or write a shader. They need to know what good looks like and how to get there through iteration with the people who build things. Frank fills that role for me now, except he doesn't need a desk at the studio.

Have a great day!

- Nick

Nick Dorra

Say hi 👋 on Linkedin​
🤝 Book a meeting via video or f2f to chat more