There are two phrases that have reliably marked every great financial bubble in modern history.
The second phrase is less discussed but equally diagnostic: “nobody knows anything.” It is what you say when the honest position is uncertainty so complete it borders on paralysis.
In May 2026, both phrases are everywhere. And remarkably, they are often coming from the same mouths.
Start with the number that puts all of it in perspective: 0.1%.
That is Bank of America’s own estimate for how much AI is currently lifting economy-wide productivity per year — published in the same report that called AI bigger than electricity and the internet combined.
The arithmetic behind the 0.1% is straightforward. AI can currently transform about 20% of all workplace tasks. Only 23% of those tasks are cost-effective to automate at today’s prices. Automated tasks save roughly 27% in labor costs. Labor is about half of all costs. Multiply it out and the theoretical ceiling today is a 0.66% gain in labor productivity — before friction, slowness, and institutional inertia compress the realized number further.
This is BofA’s own math, used to build its own bull case. Every serious argument about AI’s economic future — bull and bear alike — is an argument about whether, how fast, and at whose expense that gap closes. What follows are the two strongest versions of each side.
Mollick sounded downright Goldmanish as he addressed the New York Public Library crowd. “There’s no playbook,” he said. “We’re figuring it out. On one hand, that’s terrifying. On the other, it’s great — because that means if you create your own playbook, there’s actually a source of advantage for you in that.” Tell that to the Hollywood studios that couldn’t stop putting out bombs.
The box office bomb has its parallel in the stock-market crash, and Mollick said the fates come down to two simple but difficult questions. “The biggest picture, there’s only two questions that actually matter a lot, which is how good and how fast? How long does this exponential curve continue and at what point does it ease off and how sharp will it be? That determines everything else.”
“KPIs are the biggest enemy at this point. They force you into very bad paths in the experimentation phase,” he said. “The very nature of saying we need a 10% improvement constrains the kind of use cases that you see.”
The sharpest evidence that nobody knows anything — sharper even than the 0.1% figure — comes from Mollick’s observation about the AI companies themselves. “It’s weird that the AI companies are all now building their own consulting arms to do AI deployment. If the models are so good that you think they’re going to destroy all white-collar jobs, shouldn’t they also be able to help you deploy systems?”
The companies that built the technology and are most bullish about its capabilities cannot use that technology to answer the most basic practical question: how do you actually deploy it?
