68% of PM job postings in the last two weeks asked for AI experience.
Do you have it?
Two thirds of the market now touches AI
A quarter of the market is AI-core: the AI is the product. Another 42% expects you to be conversant. Only a third of PM roles say nothing about AI at all, and that slice shrinks every week.
The money sits on the AI side
These are not computed averages. These are the pay ranges the companies themselves posted on AI PM roles this month, as listed. Bar length shows the top of each range:
And these are not entry roles. 72% of the AI roles were Senior or above, against 58% everywhere else. This is where the promotions are.
You do not need an AI title to qualify. The postings say so.
A line from one of these job descriptions, and versions of it show up across the market:
“Shipping LLM-backed features, even side projects or prototypes, with a clear before and after.”
Read it again. A side project counts. A prototype counts. What they are screening for is whether you have built something real on a real LLM, and whether you can show what changed because of it.
Now the other half. The single most repeated line across these roles, in different words at different companies:
“Shipping LLM or agent products to real users, not demos.”
Demos do not count. A demo is something you watched get built, and every hiring manager has now seen a hundred of them. What counts is a prototype that proved a case, with a before and after.
The strongest version of that story is not a weekend toy. It is a prototype that proved a case: a real problem from your own work, sized, built, measured against a bar you set up front, and put in front of the people who own that problem. The before and after is the whole credential.
What the AI roles ask you to have done
Clustering the requirements bullets of the 93 AI-core roles, the top asks:
Notice what is not on the list. Fine-tuning appeared in exactly zero of the 255 job descriptions. Not one. The market is not asking PMs to be ML engineers. It is asking for judgment: what to automate, how to know it works, where it fails.
The capabilities behind the buzzwords
The percentages are themes. Inside them, the postings get specific. These are real requirement lines from the 93 AI-core roles:
“Building eval loops and quality bars for non-deterministic systems”
“Building evaluation, guardrail and hallucination-control loops for AI output”
“Judging where a model is good enough to deploy and where it is not”
“Product judgment on where automation should and should not take over”
“Communicating AI capability and limits clearly to non-technical stakeholders”
Strip the vocabulary and every one of these is the same test: have you built something with a model, watched it fail, decided what to do about it, and can you explain those decisions to someone who has not. None of that comes from a lecture. You can only earn it by building.
One more pattern worth knowing before an interview. The postings use the big words hundreds of times: LLM, 461 mentions. Agents, 399. Compliance and governance, 375. The precise words appear in the low tens: evals, 89. Grounding, 28. RAG, 24. The market asks the question in business language and expects your answer in builder’s language. Knowing the second vocabulary, from having built, is what reads as senior.
The interview loop is the real tell
The interview process across all 93 AI-core roles:
Nine technical rounds out of 93. They are not going to quiz you on model internals. They are going to ask what you built, why you made the calls you made, and how you knew it worked. Your AI knowledge gets tested inside a behavioral story, not on a whiteboard.
Walk in with something you built.
That is exactly what I do. We find the problem in your own work and put a number on it. You build the case and prove it to a real stakeholder. We design the AI together: what runs on code, what runs on the model, how you measure it, where it fails. You build it and deploy it, with a live URL. Then I turn into the interviewer and push until you can defend every decision out loud.
If you have AI experience already, I sharpen it. If you have none, we build it.