How to become and thrive as an AI product manager

Executive overview

Most PMs wonder whether they need a machine learning PhD to work in AI — they don't. The barrier to entry has dropped sharply: tools like Cursor, Replit, and V0 let any PM prototype real products in under an hour.

There are three types of AI PM: those building infrastructure for AI engineers, those building AI-powered consumer products, and those using AI tools to do their existing PM job better. All three require the same foundation — deep curiosity about customer problems, a working knowledge of what the models can and can't do, and a portfolio of things you've actually built.

The highest-leverage PM position right now is one who understands AI's limits, stays obsessed with the customer problem, and uses AI tooling to prototype, communicate, and amplify their influence across the whole organisation.

The three types of AI product manager

• AI platform PMs build tooling for AI engineers — observability, evaluation, debugging (e.g. Arize AI).
• AI product PMs package foundation model capabilities into consumer or enterprise experiences (e.g. ChatGPT, NotebookLM).
• AI-powered PMs use AI across their existing PM workflow — prototyping, research synthesis, design mockups, stakeholder communication.
• In the near future, most PMs will be some flavour of AI PM; AI will be as ubiquitous as the database.

Breaking into AI PM

• No degree exists for AI PM; stand out through a portfolio of things you've built, not certifications.
• Start with the foundation: Andrej Karpathy's LLM explainer on YouTube is a practical starting point.
• Use Cursor, Replit, and V0 to ship working prototypes; use Midjourney or DALL-E for visual storyboards.
• A portfolio answers two hiring questions before the interview: "Can they do the job?" and "Are they excited about this work?"
• It is easier to break in now than two years ago — ML depth is no longer a prerequisite.

What separates top 5% AI PMs

• Avoid building "ChatGPT on our own data" just because it feels intuitive — most internal chatbots have low usage.
• Ask first: is this the right problem to solve, and is a chatbot the right interface? Often it isn't.
• The same pattern is repeating with AI agents — many teams are describing a solution, not a problem.
• The better move: let foundation model companies build the agentic layer; your job is to make it seamlessly embedded in your product.
• Study successful AI products to find what makes them feel magical, then reverse-engineer the UX principles.

The Betty Crocker principle: leave knobs for the user

• Betty Crocker's "just add water" cake mix killed sales; adding eggs back restored them — customers want a degree of control.
• Waymo lets passengers control music and temperature even though the car drives itself.
• Fully automating away the user is often the wrong goal; reducing friction while preserving agency is better.
• Blue Apron is not solving hunger — it is selling the feeling of cooking; the best AI products similarly lower the barrier to creation rather than removing the human.

Finding good AI ideas inside your company

• Give your AI investment a metric: track how many experiments you're running, not just revenue impact.
• Run hackathons with specific problem statements — come in with a list of problems and see where AI actually sticks.
• Expect most ideas to fail; the Slack-bot-routing-support-tickets example failed because context was far harder than expected.
• Use tools like Gong plus long-context LLMs to synthesise hundreds of customer call transcripts — be in 100 meetings at once.
• Product teardowns (live-streaming a breakdown of NotebookLM, for example) build team literacy without eating sprint capacity.

Thriving as a long-term IC PM

Three principles: energy, waiting vs. wandering, and amplifying signal.

Energy

• Bring energy even when you don't know which direction to go — it removes friction and elevates the whole team.
• Energy doesn't mean loud; it means visibly engaged, positive, and present.
• Pair with sales, engineering, or ops on their actual work — pairing builds empathy and demonstrates you're in it with them.
• Culture flows from the PM; "make it happen" as a meeting close-out phrase can become a team value.

Waiting vs. wandering

• Executives default to waiting — "let's see what the next OpenAI model does." That is rarely the right call for a PM.
• The PM's job is to be the wanderer: go out into the unknown, gather signal, and bring back a direction.
• Feeling lost about the product roadmap is normal, not a failure — zero-to-one genuinely feels directionless.
• Keep wandering until you feel the product pulling you in a direction; that drag is the signal.

Amplifying signal with AI

• Feed customer call transcripts into long-context models to extract the most common pain points at scale.
• Use NotebookLM to condense long research inputs into something digestible.
• AI lets a single PM be present in more conversations than is physically possible — treat it as a signal amplifier.

Career mindset for IC PMs

• Stay obsessed with the customer's problem, not the technology or the job title.
• Write one page on what work means to you and one page on what life means to you; find the overlap (from Designing Your Life).
• The bar for PM impact has risen — AI tools raise what's expected — but they also raise what a single IC can deliver.
• Have fun. Curiosity and enjoyment drive faster iteration and longer staying power than discipline alone.

Recommended resources

• A Short History of Nearly Everything — Bill Bryson; puts the current AI moment in cosmic perspective.
• Designing Your Life — Burnett & Evans; practical exercises for career clarity, especially the work/life meaning overlap.
• Andrej Karpathy's LLM explainer on YouTube — foundation-level understanding of how large language models work.
• Greg Eisenberg's build-in-45-minutes video format — proof of what's now possible without deep ML knowledge.