A path, in order, from concepts to production.

Most AI education for small business is either a vendor pitch or a research summary. This is the middle: what an agent actually is, how retrieval changes what it can answer, why evaluation is the whole game, and what the operating rhythm looks like once one is live.

The sequence

• Foundations — models, context, retrieval, and why an agent differs from a chatbot.
• Design — job description, escalation policy, and the edge cases that decide quality.
• Evaluation — building a scenario suite and using failures as training input.
• Operations — the weekly loop, the metrics, and when to retire an agent.

Who it is written for

Owner-operators and ops leads, not engineers. The material assumes you can read a spreadsheet and run a business, and does not assume you can read Python. Where a technical concept genuinely matters — retrieval, evaluation, fallback — it is explained in operational terms with the consequence attached.

The one idea worth taking away

An agent's quality is set almost entirely by two things you control: the knowledge you give it and the cases you test it against. Model choice matters far less than either. Teams that internalise this ship working agents; teams that chase models keep rebuilding the same demo.