The product manager, before and after

I’ve had the title “product manager” or some variation of it since 2013. For most of that time, if you’d asked me what the job was, I’d have given you the textbook answer without thinking. Somewhere in 2024 the textbook answer stopped being true, and I didn’t notice for longer than I’d like to admit.

So here are both definitions, side by side. The old one, because it’s still what most job descriptions ask for. The new one, because it’s what the job actually is now.

Before: 2005 to roughly 2022

The canonical version: the PM sits at the intersection of business, technology, and user experience, and is responsible for deciding what to build and why. Marty Cagan gave it the four tests — valuable, usable, feasible, viable. Ben Horowitz gave it the aspiration — “CEO of the product.” Every PM course for fifteen years was a variation on those two.

In practice, the job was defined by its artifacts and its position. You wrote the PRD. You ran discovery. You owned the roadmap and prioritized the backlog. And, most of all, you translated — between engineers who could build, designers who could design, and executives who held the budget. Your authority came from being the one person who understood all three and could write it down.

Underneath all of it was one economic fact: engineering time was scarce. The PM’s core value was making sure it wasn’t wasted, and the way you did that was by specifying carefully before anyone built anything. That’s why the spec was sacred. That’s why “just try it” was a luxury.

In one sentence: the person who decides what gets built, and coordinates the people who build it.

After: 2024 onward

The decision half survived. It got harder, if anything.

The coordination half is dissolving — not because coordination stopped mattering, but because the things that made it necessary are going away. The spec was the only way to show an idea; now you can build the idea in an afternoon. Synthesis was a week of sticky notes; now it’s an hour. Engineering was too scarce to experiment with; now the experiment costs less than the meeting to approve it.

What’s replacing the coordinator is something I’d call the owner of judgment in a system where production is cheap. The post-AI PM decides which bets are worth making. They define what “good” means for a product that gives a different answer every time you ask it. They set the evals and the metrics that say whether it worked, and they are accountable when it didn’t. They show the idea instead of describing it. They edit — drafts, research, prototypes, all generated — rather than author from scratch. And they carry responsibilities that didn’t exist in the old definition: trust, safety, model cost, and what happens when the product is confidently wrong.

In one sentence: the person accountable for what gets built and whether it was right — with the taste to choose and the hands to show it.

The shift, one line at a time

• From author to editor

• From translator to builder

• From coordinator to decider

• From specifying the certain to measuring the probabilistic

• From “CEO of the product” to “owner of the bet”

Read down the left column and you get a job I did well for a decade. Read down the right and you get the job I’m retraining myself for, in public, one tool at a time.

The uncomfortable part

The first definition isn’t dead. It still describes a real job at a lot of companies, and it will for years. But it’s now the junior version. The people who write specs and relay status and run the roadmap meeting are doing work that a smaller team with better tools does faster, and everyone in the room can feel it.

The senior version is the second definition. There are fewer of those seats, and they’re not given to people who list “AI” in a skills section. They go to people who can show a bet they made, the evidence they had, the thing they built to test it, and what the number said afterward.

If you’ve been doing the first job — and I was — that’s the gap to close. Not by reading about the second job. By doing it, badly at first, and showing the work.

This is part of a series. I’m rebuilding the PM job one tool at a time and writing up the toolset it takes.