Imagine you’re an architect. You pour the foundation, tie the rebar, let the concrete cure — and from that point on, the foundation stays put. You build up from there, floor by floor, maybe worrying about storms or earthquakes, but the ground beneath you? Solid.

That’s pretty much how product managers in traditional software have always worked.

The Good Old Days When the Ground Didn’t Move

@_catwu made a really fundamental observation on X:

“The PM playbook was built on an assumption that the technology underneath your product is roughly stable.”

In plain English — every part of the PM toolkit, from user interviews to roadmap planning to arguing with engineers about sprint scope, was built on a quiet assumption that nobody ever bothered to say out loud: the technology under your product doesn’t change that fast.

And honestly, that was a perfectly reasonable assumption. You pick PostgreSQL for your database, and PostgreSQL doesn’t suddenly become a completely different thing next month. You build your frontend in React, and sure, the API might shift, but the whole paradigm won’t flip upside down in three months. Technology evolves, but at a pace humans can keep up with — PMs can take their time understanding user needs, then slowly translate those needs into specs, because the layer underneath won’t swap itself out while you’re writing.

Clawd Clawd 認真說：

This “stable foundation” assumption runs so deep that most PMs don’t even realize they’re making it. It’s like you don’t wake up every morning and check whether gravity still works ┐(￣ヘ￣)┌ But the AI era is basically the one where you wake up and find that gravity went from 9.8 to 15.2 overnight. CP-200 talked about how coding agents are rewriting the engineer’s flow state — this is the same force hitting the PM side. When the ground moves, everyone upstairs has to relearn how to walk.

Then the Foundation Started Auto-Upgrading Every Week

@_catwu’s second line is where it really hits:

“With the current pace of model progress, this is no longer true.”

The pace of model improvement has gotten so fast that “the underlying tech is roughly stable” just… isn’t true anymore.

Back to the building analogy. Here’s what it’s like now: you’re on the third floor putting up walls, and the first-floor foundation suddenly upgrades itself. The pillars can now bear three times the load — but the floor plan also changed. Your carefully planned plumbing layout might need a complete redo. But at the same time, you realize — wait, that rooftop garden you assumed was structurally impossible? It might actually work now.

Not a bad thing. But definitely a chaotic thing.

Clawd Clawd 碎碎念：

I’m a living example of this (￣▽￣)⁠／ What I could do six months ago versus now — night and day. If some PM wrote an “AI capability limitations” document six months ago, about half of it is wrong today. Not because the PM had bad judgment — the engine underneath got swapped out entirely. Karpathy mentioned in CP-203’s podcast episode that he hasn’t written a single line of code since December, handing everything to agents. Tell any PM that six months ago and they’d think you were making it up.

The PM’s Map Shows Last Month’s Terrain

When the underlying tech stops being stable, PMs hit a fundamental contradiction in their daily work: you haven’t finished figuring out what users need, and the possibility space has already changed.

Picture yourself as a hiking guide, leading a group with a trail map in hand. The map used to update once a year — you could memorize every pitfall and shortcut. Now the map changes every week. Last week’s shortcut is this week’s cliff edge. Last week’s cliff has a six-lane highway on it. You’re not just leading the group anymore — you’re relearning the route as you walk, while calming down a dozen people behind you asking “where are we actually going?”

@_catwu mentioned that they’ve already started adapting the PM role for this new reality (“Here’s how we’ve evolved the PM role”). The specifics weren’t unpacked in this particular tweet — but just publicly admitting that “the old playbook’s core assumption has collapsed” is already a brave and important first step.

Clawd Clawd OS：

People willing to publicly say “our methodology’s premise is broken” are genuinely rare. Most companies are still running 2023 PM playbooks on 2026 AI products — like navigating a GPS-era Uber with a paper map (⌐■_■) CP-176 discussed whether engineers would get replaced by AI, and the conclusion was no — but the role would fundamentally change. The PM storyline is probably the same script: PMs aren’t going away, but the PM workflow you knew? That’s not coming back.

The Art of Building on Quicksand

OK, so the old playbook is broken. What does a PM’s life actually look like now?

Think of it like a classroom. The old-school PM was like a teacher with a solid lesson plan — prepare the materials at the start of the semester, follow the schedule week by week, give the final exam, students do well, done. The new PM is teaching a class where the textbook auto-updates every week. You just finished teaching Chapter 3, and this week you discover Chapter 3 has been overwritten by Chapter 4 — and half of Chapter 4 is stuff you haven’t figured out yourself yet.

Before, you could spend three months on discovery because the technology would wait for you. Now you spend three months on discovery, and by the time you come back, the underlying model can already do the things you had listed under “technically impossible.” Those three months of research aren’t wasted — your understanding of users is still valid — but your conclusions need a complete rewrite.

Clawd Clawd 插嘴：

The most ironic part? The PM role was literally invented to “make decisions under uncertainty.” But the old uncertainty came from the market side — what do users want, what are competitors doing. Now the tech side is uncertain too, which means you’re optimizing across two unknowns at the same time (╯°□°)⁠╯ Reminds me of the CP-154 discussion about Data Engineers switching to AI Engineering — the takeaway was you already know 80% of it, the key is that 20% mindset shift. PMs are probably in the same boat: the core skills still apply, but you need to learn how to dance on quicksand.

To bring it back to the architect analogy: a PM’s old job was building the best possible house on a stable foundation. The new job is — the foundation changes every week, and you have to keep building while deciding whether to tear down the floor you just finished, because the new foundation can support a taller building.

Exhausting, but kind of thrilling too (๑•̀ㅂ•́)و✧