AI Doesn't Need More Builders. It Needs Thinkers.

I gave a talk at Pie Fi (a builder community) in Santa Cruz a couple weeks ago. A room full of builders and PMs — mostly about my lessons as a PM leader turned builder, and what to keep in mind as individuals building new products.

The line that seemed to have stuck with people wasn’t the cleverest thing I said. It was the most obvious.

“Don’t defer your intelligence to the tool.”

I said it and noticed the room went quiet for a half second. A lot of heads nodding in agreement.

I’ve been chewing on that line ever since. Not because of what it meant in the room, but because of what it means for how the job of a PM just changed underneath us.

Last week I wrote about the Connector. And that the pattern AI amplifies isn’t a title — but a posture. Someone who holds the whole picture, translates across disciplines, connects everything, and brings accumulated judgment to every decision they touch.

That article named who. This one is about what they actually do now that the tools shifted under them — specifically for PMs.

Connector. Thinker. Same person.

The PM’s job used to be translating intent into something a team could build. The PRD was usually the artifact. Ideate, spec it, plan it, hand it off, ship it.

In an AI world, that’s not the job anymore. The job is iterating fast (prototype), killing fast, deciding what’s worth building, before building it.

The PRD is still there. It’s the decision record now. The thing that captures what you decided and why, after the evaluation work (prototype, kill, start over, validate) is already done. I’ve started calling it the “Product Rationale Document” (still PRD). Completely different meaning.

After the pie-fi talk, Peter Genovese came up. Peter worked on the Yahoo! product team years ago. He’s a Google PM now, and told me: “That’s how we work now at Google. Prototypes are the new spec. The docs capture the final decision.”

The PRD now writes itself at the end of the loop — capturing the why and what we are building, not the start.

Aakash Gupta wrote about Anthropic’s approach recently. His framing stopped me: “The bottleneck was never how fast can we code. It’s how fast can you evaluate and kill. The 80% kill rate IS the quality process.”

Read that twice. The quality process isn’t the review at the end. It’s the kill rate at the beginning.

For the longest time, much of the PM role was all about the output. Status updates. Slide decks for the leadership review. Jira hygiene as a full-time job. Standup choreography. Roadmap presentations that took longer to build than the things on the roadmap.

None of that was ever the craft. It was the evidence of the craft. The part you could put in a deck because the actual work (the judgment, the tradeoffs, the no’s) were invisible. So PMs got measured on the evidence and slowly, the evidence became the job.

AI lets you produce more decks, more tickets, more status updates. More artifacts, faster. But that’s the trap.

The PMs who see the shift are using AI for the opposite reason. They’re using it to clear the artifact work out of the way so they can do the actual craft itself. Defining what’s worth building. Killing what isn’t. Saying no when it’s hard.

Those PMs are killing it (literally) right now. The other ones are running out of time.

Every PM knows the iron triangle: Time, Resources, Scope. We’ve been living with this forever. AI just collapsed two sides of the triangle (Time & Resources). Scope, is all that’s left (think about that for a moment). And focusing on the right scope was always the challenge. Which is exactly the part most PMs haven’t caught up to yet.

AI just made saying yes the easiest it’s ever been. The cost of yes used to be the team’s time. Someone had to actually build it.

Now the team is fast enough that yes feels free. It isn’t. It never was. The brake used to be engineering capacity and that brake is gone.

Saying no was always the hardest thing a PM did. The “just one more thing” that eats every roadmap. The pet feature from the VP. The ask from sales. The scope creep that felt impossible to push back on. The PMs who couldn’t say no were the ones who drowned, and products suffered from it.

AI didn’t make that easier. It made it harder. And most of the PMs I’m watching are saying yes more, not less.

Your No was always the muscle. Now it’s the only muscle that matters.

The discipline underneath is the simplest thing in the world. And the hardest.

Two lines from that same Pie Fi talk keep coming back to me. The first was the one I opened with: “don’t defer your intelligence to the tool.” The second was this: “Superpowers pointed in the wrong direction just get you lost faster.”

Both of them point at the same thing.

The reflex is “let me ask ChatGPT/Claude.” The discipline is to form your position first. Then use the tool to pressure-test it. What do you think? What’s your hypothesis? Only after that does the tool come in.

The tool sharpens what you brought to it. It cannot sharpen what you didn’t bring. That’s the difference between using AI and deferring to it. Same software, completely different person on the other side of it.

The Connector brings judgment to the tool.

The non-Connector defers judgment to the tool. That’s not a style difference. That’s the difference between leverage and replacement.

Here’s the part that should make every PM reading this sit up.

Amol Avasare runs growth at Anthropic. When Lenny Rachitsky asked him what part of PM work AI couldn’t automate yet, his answer was direct: getting six people in a room to agree.

And his bigger point, the one that matters: PM roles aren’t going away. They may actually grow.

Five engineers at Anthropic using Claude Code now ship the output of fifteen to twenty. The PM layer didn’t scale with them. One PM is effectively managing the output of a team three or four times the size it used to be.

What did Anthropic do about that?

Two things. They started hiring more PMs. And they deputized product-minded engineers to act as mini-PMs on smaller projects.

Read that again. The company everyone watches for signs that AI is replacing work, is adding thinkers, not subtracting them.

Most orgs are still staffed for a world where building was the constraint. They have plenty of builders. They’re starving for thinkers.

Here’s the thing that keeps coming back to me from that Pie Fi room.

The basic question remains: what problem are we solving? Who are we solving it for? You’ve been asking these your whole career. Your experience is the input. AI is the multiplier. You are the filter.

The role isn’t new. The recognition is.

The hardest, least visible work in the building just became the only work that still matters. The judgment. The kill. The no.

If you’re a Connector reading this, your job didn’t shrink. It got harder. It got more important.

The thinking job was always the job. AI just made it the only job.

The question is whether the generation behind you will know how to do this work at all (Good topic for the next article.)