Earlier this year, I spoke with a fintech executive who described his team as aggressively AI-forward. One rule he enforced has stayed with me: he told every product manager on his team never to walk into a review with an AI-generated strategy deck that lacked substantive thinking behind it. The underlying assumption was that telling the difference between real insight and hollow polish would be obvious.
It’s the irony of modern knowledge work: companies mandate AI for everyday execution, only to immediately caution their teams against leaning on it. We picture these decks as finished, beautifully formatted strategy proposals filled with recommendations that are either unworkable or, worse, reasonable-sounding on the surface yet impossible for the PM to defend when pressed.
Researchers at Stanford and BetterUp Labs coined a term for work that looks pristine on paper but lacks substance: “workslop.” Nearly 40% of full-time US employees said they had received workslop in the past month. And the suspicion runs both ways: in one survey this year, 55% of employees said they had received workslop from a manager or supervisor.
As Shopify CEO Tobi Lütke has noted, learning to use AI well is an “unobvious skill.” Meta and other companies now evaluate teams on how effectively they leverage AI to drive outcomes. What most organizations haven’t worked out is how to judge what constitutes an “effective” AI workflow.
The Middle Is Hollowing Out
Before generative AI, a PM tackling a complex roadmap or positioning question spent days synthesizing user research, data, and constraints. Most output landed safely in the middle: competent, defensible, and unremarkable.
That middle is rapidly disappearing. Faced with organizational pressure for speed, a PM now takes one of two paths:
The Speed-at-All-Costs Path: The PM uses AI to churn out a 50-page PRD or a polished strategy deck in 20 minutes. It is fast, and it is worse than what they would have built alone: disconnected from core technical debt, ignorant of telemetry data, and riddled with unexamined edge cases. If an organization rewards raw velocity without rigorous quality checks, this is the default result.
The High-Leverage Path: The PM uses AI as an adversarial sparring partner, simulating cross-functional pushback, stress-testing business cases, and vetting user constraints before taking it to the team, a step AI can prepare for but not replace.
The part leadership routinely misses is that the second path takes just as much cognitive effort, and often just as much calendar time, as the old process. When placed side-by-side with the instant gratification of a workslop deck, careful AI work looks slow. Leaders who evaluate AI purely on speed end up accidentally penalizing depth.
What Separates the Outperformers
In 2005, an open chess tournament allowed any combination of human and machine. Most expected a purpose-built supercomputer to dominate. Instead, the tournament was won by two casual amateurs running off-the-shelf chess programs on ordinary computers. Their secret wasn’t raw processing power; it was method. As Garry Kasparov later observed, a weak player with a machine and a superior process beats both the strongest computer and a strong player with a flawed method.
Two decades later, the same pattern is showing up in product organizations. A KPMG and University of Texas study published in the Harvard Business Review tracked professionals against an AI baseline: roughly half outperformed it, a quarter matched it, and a quarter fell below it. The outperformers looked identical on paper; what set them apart was how they framed problems, grounded inputs in domain expertise, and iteratively refined the output over multiple loops.
Judging the Process, Not Just the Output
A recent Wharton survey found that over 70% of managers have shifted from creating outputs to managing AI-generated outputs. The hard skill for product leaders is no longer teaching teams how to prompt; it’s learning how to interrogate what holds up.
For most of the history of product management, a polished deck didn’t prove the thinking behind it, but it said something about the PM: they cared, and they sweated the details. That usually went hand in hand with good thinking, and it made pressure-testing easier, because the review could start with the argument instead of the basics. AI made polish free, and that signal went out the door.
That changes what a Director or VP should do in a review. Asking whether AI touched the deck is the wrong question. Instead, the review criteria must return to the same age-old questions that separate rigorous product management from hand-waving:
Why this specific trade-off?
What assumptions did you rule out?
What did you stress test, and how can we feel confident this is the right answer?
When a PM uses AI to pressure-test their own thinking, those answers hold up.
The executive assumed hollow polish would be easy to spot, and in one sense he was right. What gives a deck away isn’t AI. It’s the absence of thinking.