problem hacker #28
The Bubble Whisperers Are Wrong
There’s a vibe shift happening.
It’s not loud. It’s not the FT front page.
It’s in Slack threads.
In the way someone lowers their voice slightly before saying “I’m not sure the AI thing is quite what everyone says.”
And yes — for every hushed doom post, there are a hundred LinkedIn grifters promising $600 a day with a Mac Mini. That’s embarrassing and it deserves every eye-roll it gets.
But the backlash is making a dangerous move. It’s using the grift as evidence against the technology.
That’s a category error.
And it’s going to cost people who act on it.
The problem
Two numbers are doing most of the damage right now.
- 80% of companies report no productivity gains from AI. (NBER, 6,000 executives, 2026)
- Only 3% of AI users are paying customers. (Menlo Ventures, 2025)
Cited together, they sound like a verdict. They’re not.
They’re a misreading of where we are in a cycle that has played out — identically — three times before.
We’ve seen this film play out before
Robert Solow said in 1987:
“You can see the computer age everywhere but in the productivity statistics.”
A few years later, the 1990s productivity boom arrived and everyone who called the tech bubble in 1989 looked foolish.
Before computers, electricity.
Factories began electrifying in the 1880s.
For thirty years — thirty years — no meaningful productivity gain showed up in the macro data.
Then, just before 1920, it all arrived at once.
Electrification accounted for half of all US manufacturing productivity growth in the 1920s.
What changed? Not the electricity. The factory floors.
Managers had spent three decades swapping out steam engines for electric motors — same layout, same logic, different power source.
The gains only came when they tore the factory down and rebuilt it around the logic of electricity.
We are in the steam-engine-to-electric-motor phase of AI right now.
The macro data is quiet because almost no one has rebuilt their factory floor yet.
The numbers people aren’t using
The 80% figure is doing something sneaky.
Those same executives reported using AI for an average of 1.5 hours per week.
A quarter didn’t use it at all.
You cannot measure a productivity revolution in 90 minutes a week.
Meanwhile the micro studies tell a different story.
P&G found individuals using AI performing as well as a team of two.
BCG consultants completed tasks 18% faster.
The macro data is averaging deep adoption with CEOs who opened ChatGPT twice in November.
On the paying-user stat: generative AI reached 1.7 billion users in 2.5 years.
Facebook took five years to reach meaningful revenue.
Twitter had 200 million users before it hit $1 billion. These weren’t bubble signals — they were early monetisation curves.
The gap between “uses it” and “pays for it” has always preceded the inflection, not signalled failure.
And one more thing the whispers ignore: workers with AI skills now command a 56% wage premium over peers without them — up from 25% just a year ago.
Markets don’t pay premiums for things that don’t work.
The hack
The problem isn’t that AI doesn’t work.
The problem is that you’re measuring it wrong.
Most organisations are tracking AI like a cost centre — licence fees in, productivity hours out.
That’s the electric motor swap. It misses the point entirely.
The question to ask instead is: where in our operation is AI already changing the shape of a decision?
Not time saved. Not tasks automated. Decisions changed.
That’s where the factory floor redesign begins.
That’s where the productivity data will eventually show up. And the organisations asking that question now are building an advantage that won’t be visible in anyone else’s numbers for another five years.
The bubble isn’t in the technology.
It’s in the assumption that change this deep happens in two years.
The Problem Hacker identifies problems businesses don’t know they have.