When All the Roles Merge: Reflections on PM in the AI Age

Reading Cat Wu’s post Product management on the AI exponential — published on the Anthropic Claude blog — felt like reading my own story.

One line stopped me cold:

“All of the roles are merging. PMs are doing some engineering work, engineers are doing PM work, designers are PMing and also landing code.”

Two years ago, I joined my current company and built an AI knowledge management app called Knowledge Assistant. With AI, I did everything: prototyping, UX, user research, evaluation frameworks. My section chief and deputy director both called me “irreplaceable” at different points, which remains the nicest professional compliment I’ve ever received.

But nobody knew how to define my role.

So everyone, inside and outside the team, started calling me the product owner.

When the Boundaries of PM Begin to Dissolve

Cat Wu explains that traditional product management rests on a foundational assumption: what’s technologically possible at the start of a project is roughly what’s possible at the end. PMs were trained to gather information upfront, make long-horizon bets, and execute against a plan.

Exponentially improving AI has broken that assumption entirely.

What a model can’t do today, the next version might handle in one shot three months from now. In this environment, six-month roadmaps aren’t just inefficient. They’re dangerous. The new rhythm, as Cat Wu frames it: rapid experimentation, consistent shipping, and doubling down on what works.

Prototypes can be built in an afternoon. Wrong bets are cheap. The old “write the spec first” discipline gives way to: build it, let someone use it, watch what actually sticks, then iterate.

This maps directly to my experience in enterprise AI. Model capabilities don’t improve linearly — they jump. The evaluation framework you build this month may need major revision when the next model drops. Waiting for a complete spec is almost always too slow.

Peter Deng’s Five PM Archetypes — AI Makes You All of Them

Cat Wu’s article brought to mind an insight from Peter Deng, former VP of Product at OpenAI. He categorizes PMs into five archetypes:

1. Consumer PM — obsessed with detail and visual craft
2. Growth PM — data-driven in everything
3. Business/GM PM — focused on business models and margins
4. Platform PM — builds internal tools for team leverage
5. Research PM — bridges frontier technology with market needs

Historically, one person went deep on one type. You built a complete product team by assembling all five — Deng’s “Avengers” analogy.

AI has rewritten that rule. And this connects directly back to what Cat Wu describes as the new PM velocity.

When one person can spin up a prototype in an afternoon, design a UX with AI assistance, run systematic evals, and validate growth hypotheses with data — that person effectively becomes the entire Avengers team. This is the natural extension of what Cat Wu calls “role merging,” and it’s what I’ve lived daily for the past two years. About a year ago, my role finally crystallized: AI Product Manager — a new archetype that uses AI to span prototyping, UX, evaluation, and research, often as a single person. Basically doing everything.

Building for a Future That Doesn’t Exist Yet

Another line from Cat Wu that stuck with me:

“You need to build products that don’t yet fully work, so you’re ready when the next model closes the gap.”

This is a manifesto against perfectionism, and the most practical product philosophy for the AI era. You can’t wait for the technology to be perfect before shipping, because by the time you’re ready, the technology will have moved somewhere else entirely.

In practice, this means having the courage to ship something “not quite good enough,” then iterate rapidly as model capabilities improve. Yes, you might rewrite everything in three months — but you’ll have accumulated irreplaceable user feedback and real-world insight that no spec document can match.

When Everyone Is Accountable, Growth Has No Ceiling

If AI enables everyone to cross role boundaries, how much do titles and job descriptions still matter?

My answer: the less titles matter, the more accountability matters.

AI empowerment is equally available to every member of an AI-native team. Engineers can do PM work with AI. Designers can do engineering work with AI. PMs can do research work with AI. Remove the constraint of “my role is X,” and individual growth becomes genuinely limitless.

But the prerequisite is that everyone must treat the product’s success or failure as their own.

Not just executing their KPIs. Genuinely holding the user’s value in mind, moving proactively, covering for each other. Peter Deng wants an Avengers team where everyone is distinct and opinionated. Cat Wu points to mission alignment as Anthropic’s friction-reducing superpower. I’d add a third ingredient:

When everyone treats the product’s outcome as their personal outcome, the team becomes unstoppable.

---

Cat Wu’s article confirms what I’ve felt for two years: we’re not just watching tools get upgraded. We’re watching the entire practice of product work get redefined. The person who built Knowledge Assistant two years ago, doing everything without a clear title, now has a clearer name. But whatever the title, the drive is unchanged: use AI to make the impossible possible, then put that possibility in more people’s hands.

💬 Further Reading: The AI PM Mindset: Peter Deng’s Philosophy at Uber