Read in the order
the job actually happens.

Four tracks, named after the situation you are in rather than the topic. Each piece ends with something you can use in a meeting, and every factual claim records the page it was checked against.

8 of 21 pieces written About this library

The planned count is an editorial intention, not a schedule. We show the gap for the same reason the certification pages show how many questions are authored: a half-finished path presented as a complete one is worse than an honest one.

Good judgment starts with a little perspective.

In the room

Someone is pitching you an AI project, or demonstrating one. What to ask, what to discount, and how to tell a real capability from a good demo.

6 of 6 written

Step 1Eleven questions to ask when someone pitches you an AI projectYou don't need to know how it works to run a good review. You need to know which questions have uncomfortable answers.6 min read · reviewed 2026-07-22 Step 2What a demo can't show youA demo is a performance of the best case. Three requests turn it into evidence, and all of them take under a minute.8 min read · reviewed 2026-07-29 Step 3How to read an accuracy claim"94% accurate" is not a fact about a system. It's a fact about a test somebody designed, and the design is where the interesting part lives.8 min read · reviewed 2026-07-29 Step 4Running a two-week bake-off that ends in a decisionMost pilots end in a conversation about whether it felt better. Write the acceptance criteria before you see a single result and that can't happen.9 min read · reviewed 2026-07-29 Step 5Build, buy, or extend — and when to do none of themMicrosoft's own blueprint makes this a testable decision, which tells you it's a real one. The framing that helps is what you'd be sad to have someone else own.8 min read · reviewed 2026-07-29 Step 6Deciding not to proceedThe hardest meeting in the track. How to say no in a way that's about the evidence rather than about enthusiasm, and how to leave the door open honestly.8 min read · reviewed 2026-07-29

Budgeting

What these systems cost, why the first estimate is usually wrong, and how to build a forecast you can defend to finance.

1 of 5 written

Step 1What an AI feature actually costs, and how to forecast itTokens are the unit you get billed in, and almost nobody explains them in terms a budget holder can use. Here's the arithmetic, with real prices.8 min read · reviewed 2026-07-24

Fundamentals

The small amount of mechanism you actually need in order to make good decisions. No maths, no code, nothing you will not use.

1 of 5 written

Step 1Why it makes things up, and what you can actually do about itHallucination isn't a bug someone forgot to fix. Understanding why it happens tells you which controls work and which are theater.7 min read · reviewed 2026-07-23

Risk and governance

What to require before something ships, what to disclose, and what to do when it goes wrong in public.

0 of 5 written

Not written yet. This track is planned rather than published. If it is the one you want next, say so and it moves up.

On its own

Pieces that do not belong to a track, and would be worse for being forced into one.

CertificationsThe three AI certifications worth having if you don't write codeMicrosoft, Google, and AWS each publish a foundational AI certification aimed at business audiences. Here's what each one actually covers, what it costs you, and the order I'd take them in.9 min read · reviewed 2026-07-25

All 9 articles, newest first →Certifications →

Tell us which track to finish.

One email when a piece lands. Nothing else, and one line to us stops it.