Michael Ruescher

I'm building Bitter .

After a lot of building with AI agents, my read is simple: the demo comes together quickly now. The real work is everything around it.

The model needs a repo, credentials, logs, deploys, memory, permissions, and a person who stays responsible for the result. Bitter is a software factory for that.

The bitter lesson is the background assumption. General methods that scale with compute keep winning. I want the systems around them to get better as the models get stronger.

If you're working through the same problem, I'd like to compare notes.