Most teams trained the PMs and sped up the paperwork.

The job did not get replaced.

You already forecast. What can be true in a year or two, what has to start this quarter, which dependency is fake. AI changed the object of that forecast: what a prototype costs, how a model fails, what gets cheap before the roadmap does.

In practice, if you have not watched a model assign probability to the next token, or seen a 4-bit quant miss a logic question with full confidence, you are calling those shots from a brochure. The certificate does not fix that.

A scale I actually use.

Five levels. A little overwrought. Roughly what I see after an “AI rollout.” Judge them by what this person’s work changes for other people, not by how clever the prompts look. Most people who finished the training are still at 1.

1. 1

   Assisted

   Same workflow. Drafts come back faster. Summaries, rewrites, notes. This is where the course ends.

2. 2

   Fluent

   You pick the model on purpose and you build the context instead of dumping the ticket into the box. You are faster. The org is not.

3. 3

   Builder

   You leave the meeting with something running. A prototype, a pipeline, an eval set. Engineering can argue with that instead of a slide.

4. 4

   Orchestrator

   Someone else can install what you made. Until then it is personal productivity with better tools.

5. 5

   Multiplier

   Discovery and ship criteria assume a prototype is cheap. The way of working still holds if you leave.

Most buildings are a pile of 1s, a few 3s, and no 5s. Four and five are usually a permission problem. Someone with authority has to let a PM build.

Before a customer sees it.

Speeding up go-to-market only counts if you can still answer these. If a team cannot fail the test, it is not a test.

• Grounding

  If the answer has no source, it does not go to someone whose decision costs money.

• Data

  You can name which data hits which model, under what agreement. “It’s just a prompt” is how data leaves the building.

• Owner

  A named person owns the output that reaches a customer. No name, no ship.

• Measurement

  A baseline, and a number that could prove you were wrong. Enthusiasm does not count.

The book and the labs.

I wrote the book so the chapters were not just claims. The labs are how you check.