The Greyquill Blog
Notes on making AI actually work in the enterprise. Foundation first: structure and clean data, then governed, domain-specific agents on top.
A frontier-class model on a 128GB machine
Ox Alpha topped OpenRouter before anyone knew who built it. It turned out to be an open-weight model you can run inside your own perimeter, which matters more to a regulated enterprise than any benchmark.
A hallucination is the model working correctly
A model is random by design. The reliability comes from the harness around it: deterministic code that calls the model like a function and checks every answer it returns.
The four lines of code that keep our AI chat honest
We put a chat on our website that answers from what we have published and nothing else. Building it taught us that reliable AI is mostly a content and harness problem.
AI climbed the first half of Bloom's ladder. The rest is human work.
Machines now remember, understand, and apply. Judgment and creation still need us. That should shape how we educate the next generation, and how we build the tools they grow up inside.