The wave of AI products being built right now is unlike anything I've seen in my career. Every week there's a new capability unlock, a new API, a new way to build something that was impossible six months ago.
But here's the thing nobody tells you: most of the classic PM skills transfer perfectly. And the gap between a great PM and a great AI PM is smaller than the hype suggests — if you're willing to learn the right things.
What transfers directly
User empathy is still everything. The most common failure mode I see in AI products isn't the model — it's that nobody properly understood what the user actually needed. A 95% accurate model solving the wrong problem is still a bad product.
Ruthless prioritization. With AI, the temptation to do everything at once is even stronger. The model can do anything. That doesn't mean you should build anything.
Clear thinking about success metrics. What does "good" look like? For AI features this is genuinely hard — but the habit of forcing clarity before you build is the same.
What you need to learn
Develop intuition for model capabilities and limits. You don't need to understand backpropagation. But you do need to know: what kinds of prompts fail? Where does the model hallucinate? What's the latency and cost profile? This comes from spending time actually using the API, not just reading the docs.
Think probabilistically. Traditional software is deterministic. AI is not. Your mental model of "user does X → system does Y" breaks. You need to think in distributions: "for this type of input, the system usually does Y, sometimes does Z, rarely does W."
Understand evaluation. How do you know if version 2 of your prompt is better than version 1? This is an unsolved problem in many teams. Getting comfortable with evals — even simple ones — is a superpower.
The honest truth
The teams shipping the best AI products right now aren't doing anything magical. They're just very clear on the problem, very honest about what the model can and can't do today, and very fast at iterating.
The best preparation? Build something. Even a small side project using an LLM API will teach you more about AI product work than any course.
If you're working on AI products and want to compare notes, reach out.