Writing / Field guide
How the parts of an AI system fit together.
I start with the model and the agent, then follow them into real data, permissions, and work.
System map / 01 to 03
MODEL → HARNESS → WORK
01 / MODEL INFERENCE
What is an LLM?
How a language model is trained, what happens during a response, and why fluent text still needs evidence.
Open the LLM explainer
MODEL / WEIGHTS / NEXT TOKEN01$ agent run inbox.reply
02goal = "prepare a response"
03state = harness.load()
04next = model.infer(state)
05result = tools.execute(next)
06harness.save(result)▌
ONE CALL → ONE ACTION → NEW CONTEXT
02 / AGENT HARNESS
What is an AI agent?
Where an agent's goal and continuity live, and what it would actually take for one to go rogue.
Open the agent explainer
03 / IN THE ENTERPRISE
Where the system meets work
Real data, permissions, processes, and people change what a useful answer requires.
Explore the use-case questionsUse cases / Field notes
Where the design choices show up
Evidence can be incomplete, people can have different access, and a plausible addition can change the shape of a product. These notes start from the problems I have encountered in the work.
Field notes are being revised. New use cases will appear here as they are ready.