Foundations

The eight jobs AI does reliably for a company of 5 to 200 people, the jobs it fails at, and a plain test for telling the two apart before you spend a dollar.

A large language model explained with no technical background: what it is, what it is genuinely good at, why it invents facts, and why it is not a database or a search engine.

Token, context window, RAG, agent, embedding, fine-tuning, hallucination, MCP, vector database and forty more, each defined without jargon and with a note on why an owner should care.

Per-token pricing turned into dollars per thousand customer conversations, subscriptions versus usage, the small server, and the two hidden costs nobody puts in the proposal.

The decision rules for choosing between a SaaS AI subscription and a system built for you, the lock-in traps that catch owners a year later, and the assembled-from-open-tools path that usually wins.

The pilot with no owner, the homepage chatbot with nothing behind it, automating a process that does not exist, buying the demo, running with no logs. Ten ways money gets wasted on AI, and the fix for each.

Data in one place, a written process, an internal owner, a rule for what the AI may and may not do, a budget, a success measure. Twenty checks, grouped into seven areas, that decide whether an AI project sticks.

Month one: one process, one owner, one measure. Month two: the second automation and the logs. Month three: the first AI step. What to buy, what to build, and the moment a developer becomes worth it.