AI is now the product manager's job, not the engineer's

We counted the skills employers list on every open product role on our index — 3,504 live PM, PO and product-lead listings at European companies. Of those, 24% name at least one AI skill. When we ran the same count on engineering roles last week, the figure was 19%. The people most often asked to put AI on their CV in Europe right now aren't the engineers who build it — they're the product managers who decide what gets built. Here is who that lands on, and why the thought leaders think the number is measuring the wrong thing.

For the job seeker: AI fluency is becoming the price of a senior product seat

Broken down by level, AI shows up in just 11.8% of mid-level product listings — then climbs steeply: 28.0% at senior, 27.5% at lead, 37.6% at staff and 44.8% at principal. A principal PM role is nearly *four times* as likely to require AI fluency as a mid-level one. (Junior roles are a quirky exception at 22.6% — a small pool skewed by dedicated "AI PM" graduate openings.) Employers aren't hiring one junior specialist to "own AI." They expect their most senior product people to lead the conversation.

For the vacancy: employers are writing "use AI," not "build AI"

The specific asks make the intent clear. Beyond broad "AI/ML fluency" (99 listings), the roles lean on the applied stack a product manager can actually wield: LLMs (82 listings), generative AI (56), agentic AI and AI agents (44), prompt engineering (22) and RAG (19). Almost nobody is asking a PM to train a model. They want someone who can scope an LLM feature, judge whether it works, reason about its cost and guardrails, and prototype with the tools directly. The skill being priced into the vacancy is *product judgment applied to AI*, not the maths underneath it.

For the job market: AI has stopped being a specialism

You might expect AI hiring to be concentrated at Europe's frontier startups — the venture-backed "unicorns." It isn't. Product roles at Europe's unicorns name an AI skill 27.4% of the time; across the rest of the market it's 23.8% — close enough that AI fluency has stopped being a signal of where a company sits on the funding ladder. And while the language is everywhere, roles built *entirely* around AI are still a minority: 6.7% of listings put AI in the job title itself. In other words, "AI" is no longer a job you hire for — it's a competency the market now assumes across product work, whoever the employer is.

For the employer: the bottleneck moved from output to judgment

Here is the tension in the numbers. Employers are loading AI into the job spec because AI makes teams faster — but the people who think hardest about product argue that speed is exactly the trap. Writing last month, Melissa Perri put it bluntly: for years the one thing that slowed the "build trap" was effort, because building was expensive and teams had to be at least a little selective. "AI just removed that brake." Her firm's survey found the measurable wins from AI landing where engineers build, not where customers feel it: "the work got faster without getting wiser."

Marty Cagan reached the same place from a different door. In *The AI Productivity Paradox* (23 July 2026) he warns that most teams are pointing AI at delivery, not discovery — that the tooling "is designed to deliver output, rather than outcomes." He quotes Hilary Gridley: "it's never been faster to build, which means it's never been easier to run 10 times faster in the wrong direction," and Chip Huyen: "the hardest part remains knowing what to build." That is precisely the part AI does *not* do for you.

Read against our data, the message to employers is uncomfortable. The AI skill worth hiring for isn't the one that ships more features — that part is nearly free now. It's the discipline to say no to the wrong work once building it stops being the constraint. If a job spec asks for "AI" but means "produces output faster," it is buying the paradox. If it means "makes sharper calls about what's worth building at all," it is buying the thing that's actually scarce.

What this means if you're a product manager in Europe

If you're a senior PM in Europe, having a real point of view on shipping AI features — scoping, evaluation, guardrails, cost, and knowing when *not* to build one — is fast becoming part of the job description, not a bonus. If you're earlier in your career, the AI door is narrower at your level, and the fastest route in is the applied stack (LLMs, agents, prompting, retrieval) paired with the product fundamentals employers already expect. Either way, the fluency now on 24% of specs is table stakes for the roles above it. You can see which of the live roles on workinproduct ask for it, or jump straight to the AI product roles across Europe — and read the spec for what it's really testing: output, or outcomes.

*Method: skill requirements across 3,504 active product listings (PM / PO / product lead) on workinproduct.eu, enriched from the original job postings; "AI" = any listing naming an AI skill (AI/ML, LLMs, generative or agentic AI, prompt engineering, RAG, and the like) in its skills field. The engineering comparison uses the same method on 7,422 engineering listings. Unicorn split uses our flagged European unicorns vs all other companies. Quotes: Melissa Perri (LinkedIn, June 2026) and Marty Cagan, "The AI Productivity Paradox," svpg.com, 23 July 2026. Snapshot 31 July 2026.*