Lessons · Batch 01
On Harnesses
One worked example runs through this batch: planning a family's meals for the week. Larder starts from an empty folder and builds the harness around one agent: the plain files it reads at the start of every session, one program it runs, and the gates where it stops for you. Ontology writes down the small world that harness works in, a model of the kitchen a script can check, so the answer stops changing between sessions. Both are in Claude Code, and neither requires the other.
- Larder
Build a small assistant for dinner out of one agent and a folder of plain files, and learn the shape that transfers to most of the other things you keep doing by hand: when the answer is wrong, correct the file, not the answer.
- Ontology
Operationalize an ontology with a language model. Give the agent a small model of the kitchen instead of a list of what is in it, so that three sizes of model give the same correct answer, and check the kitchen model with a script, because it is only worth having once something checks it.
New to agents? Read AI harnesses are for everyone first. Larder estimates what a file costs an agent to read; Representation measures tokens, cost and correctness on one Apollo 11 question as the same facts change shape, over recorded runs or live with your own key.