The thing Nikhyl Singhal said that stuck wasn't that half of PMs are in trouble. It was how he described the ones who aren't yet: "smiling exhaustion." Outwardly fine. Internally running on fumes.
That phrase does a lot of work. It describes something precise — a performance of competence sustained past the point where the role still makes sense.
A few months ago, we watched something happen across several of the teams that use Hamster. Engineers were shipping 4x faster. Founders were closing the loop from customer call to spec in a single session. But the people sitting between discovery and delivery — the PMs, the product leads — were more buried than before, not less.
The workload went up. The mandate didn't clarify.
That pattern has a name: role stretch. It's what happens when a job description stays the same while the operational reality around it shifts. You don't get a new charter. You just get more surface area. More tools to evaluate, more PRDs to write, more stakeholders to align — layered on top of everything that was already there.
Role stretch is different from being overloaded. Overload is too much of the same work. Stretch is too much of the wrong work, because nobody has decided what the right work is now.
The default story is that AI is automating PM work — user research synthesis, PRD generation, competitive analysis, roadmap formatting. And that's true, technically. All of those tasks compress with the right tools.
But compressing tasks that used to take three hours into twenty minutes doesn't free up two hours and forty minutes for strategic thinking. In practice, it frees up two hours and forty minutes for more tasks of the same kind. The PM becomes a faster version of their old self, running at higher volume, not operating at a different level.
What AI actually changed for PMs is the abstraction floor. The work that used to justify the role — translating customer needs into specs, coordinating between design and engineering, managing the discovery-to-delivery handoff — is becoming faster and cheaper for everyone on the team to do, not just PMs.
Founders who would never have written a structured brief two years ago are now doing it in twenty minutes with an AI coding tool as their thinking partner. Engineers who used to wait for a PRD are sketching interaction designs directly in code. The specialized translation layer that PMs occupied is thinner now, not gone.
That's the identity crisis. The PM isn't being replaced by a model. They're being squeezed from both ends by people who no longer need them to perform the functions that used to define the role.
Here's what "smiling exhaustion" actually looks like in practice.
The PM who's performing the role writes detailed specs that nobody fully reads, because engineers have enough context from the brief and the conversation. They run ceremonies that once served coordination functions that the tooling now handles. They produce artifacts — roadmaps, status updates, priority matrices — optimized to demonstrate that product thinking is happening, not to change what gets built.
The work looks like PM work. It just isn't doing what PM work used to do.
The PM who's operating differently asks a harder question first: given what AI now makes fast and cheap for everyone, what does only my attention actually move? Not what does my role say I own. What, if I didn't attend to it, would actually go wrong?
That question cuts differently than a job description. It points at decisions nobody else is positioned to make — the tradeoffs that require holding customer context and engineering constraint and business direction simultaneously, at the moment they need to be resolved, not reconstructed afterward.
The shift isn't from "PM who writes specs" to "AI-assisted PM who writes specs faster." It's from spec-producer to decision-owner. The artifact changes. The accountability doesn't.
The PMs who are figuring this out aren't adding new AI skills to their existing practice. They're cutting the work that looks like PM work but isn't producing decisions anymore.
That's a harder change than it sounds. Busyness is a defense mechanism. If your calendar is full of stakeholder syncs and your Notion is dense with artifacts, it's uncomfortable to ask which of those things would break something if you stopped doing them. The answer is usually fewer than you'd expect.
The PMs who make it through the next two years will not be the ones who learned to use AI tools fastest. They'll be the ones who got specific about what their attention is for.
That means fewer meetings that exist to demonstrate alignment rather than create it. Fewer documents that perform strategic thinking rather than record a real decision. More time sitting in the exact moment where customer input, technical constraint, and product direction collide — because that's the moment that has always needed a human, and still does.
The role doesn't need a new name. It needs a different theory of what it's for.
Nikhyl's prediction — companies shedding 30,000 PM roles to rehire 8,000 AI-first ones — is worth sitting with not as a threat but as a signal about what kind of PM work is durable.
The 8,000 aren't doing ten times more of what the 30,000 were doing. They're doing a different thing: holding the clarity that AI can't generate for itself. Not writing the spec, but knowing which version of the spec reflects what customers actually need versus what sounds reasonable in a planning meeting. Not generating options, but making the call when two viable directions have genuinely different implications for the product.
That judgment doesn't compress. It requires accumulated context — about the customers, the constraints, the direction — combined with the willingness to make a call and own the downstream consequences.
What AI made clear is that everything else was scaffolding around that judgment. Some of the scaffolding was useful. A lot of it was just the overhead of operating without the tools we now have.
Smiling exhaustion is what happens when the scaffolding stays up after the judgment underneath it has been crowded out. The fix isn't learning to smile less or work harder. It's taking down enough scaffolding to find out what the job actually is now.
That's not a comfortable process. But it's the only one that works.