A running list of what we think is worth sharing.

Software that can take a goal, choose the next steps, and use tools to carry them out.

The trained system that reads, reasons, and writes. It is the engine inside the agent.

The instructions and context given to the model. Most of the quality comes from what you put in front of it.

Everything the model can see while it works: the request, files, history, rules, and source material.

The small chunks of text a model reads and writes. Length, cost, and memory pressure are counted in tokens.

A capability the agent can call: read a file, search the web, run code, send a message, or update a record.

Saved context that carries across sessions, so work starts with the person, business, and rules already in view.

A reusable way to do one kind of work well, with the steps, standards, and sources already packaged.

A result produced by an agent: an email, invoice, brief, document, build, or decision record.