AI Made the Bad Decision Faster ★
Speed is a multiplier. It multiplies good judgment and bad judgment with equal indifference.
The conversation around AI in product has been stuck on the upside: how fast you can write a spec, ship a prototype, iterate on a design. All of that is real. But the downside multiplies too. The PM who used to spend two weeks writing a mediocre requirements doc now writes it in twenty minutes. The bad idea that might have died in a planning meeting has code attached to it by Tuesday.
I’ve watched this happen. Teams shipping features in days that used to take quarters. Some of those features were good. Most were features that should have been killed earlier, but the whole pipeline was faster now, so the bad idea had less time to die before it had an owner, a ticket, and a launch date.
The hard part was never writing the spec. It was deciding whether the feature was worth building at all. That judgment doesn’t speed up because the typing is faster. If anything, it gets harder to exercise when everything downstream of the decision gets cheap.
The companies getting this right are treating AI as an amplifier for product intuition that already existed. Not “AI-first” as an org-chart rebrand, but as a genuine upgrade to how individuals who already had judgment can operate. The PMs who were already rigorous, who could hold the full system in their head and anticipate what to kill before it shipped, are now absurdly productive. The ones who weren’t are failing faster, which at least has the virtue of being informative.
There’s a version of this that ends well. The floor on prototyping dropped far enough that you can validate assumptions in hours that used to take sprints. But validating faster only helps if you’re asking the right question. “Can we build this” has always had an obvious answer. AI just made it cheaper to prove it.
The skill gap isn’t prompting. It never was. It’s knowing what not to build. That part doesn’t get faster when the tools do, and nobody’s building a model to tell you.