now

Blender sculpting research, with honest results

I am testing how far an agent can get with Blender sculpting: what it can block out, what it can refine, and where a person has to take over. Each test is written up with the method, the time it took and a plain verdict on whether the result is usable.

Negative results go in the ledger with their numbers. The research is not finished, and I am not claiming a finished method. If a test shows that the approach fails, that failure is the result I will publish.

Building agent-driven 3D

I am building 3D scenes with AI agents driving Blender and procedural code. The work is at prototype stage. Some scenes render the way I want and some do not, and I keep both, because the ones that fail say as much as the ones that work.

What I can show is the scene, the code or the Blender steps that made it, and a render I have looked at with my own eyes. What I cannot yet show is rigged character animation at volume, which is not proven. Anything on this site about 3D names what was made and how far it got.

Learning AI engineering from the ground up

I am learning AI engineering from the first principles: how a model is called, what a context window holds, how a prompt becomes a tool call, and where the failures show up. I am doing it in order and in public. Each step gets a short note with what I tried, what broke and what I would check next time.

The notes are not a course and not a promise about pace. They are a record. They sit next to the projects that use what I learned, so a claim about a technique can be traced to the work that tested it.