Fast AI Hits Real Bottlenecks More Quickly
Cowen puts a rough number on the slow part and points to fax machines and school software.

The a16z Show · Alex Imas & Tyler Cowen | AI Is Moving Fast. Why Isn’t the Economy?In a podcast interview recorded November 11, 2026, economist Tyler Cowen was asked whether recent AI advances had changed his economic outlook. He answered with a warning about bottlenecks.
"often the faster that some things move, that just means you hit the real bottlenecks more quickly," he said.
Asked whether recent developments had changed his views, he said, "So I don't update that frequently." Faster technology, in his argument, exposes limits elsewhere rather than removing them.
He added longer context. Since August, he said, he had changed his views a fair amount because of math progress in areas with few human bottlenecks, where a clearly better model could appear within something like three weeks. But over the last week or two, after a meter report, he was slightly, very slightly more bearish.
His main example was the large, slow moving part of the economy. Government, nonprofits, highly regulated parts of health care, higher education, and general schooling make up a huge share of activity, he said.
"You could say like, I don't know, 40% of the US economy is a kind of nightmare," he said. "In Europe, it could be higher, like in France," Cowen added, and he said those sectors might need to operate that way for other reasons.
Cowen said "regulation is not the problem." "It's a problem," he said. "But just inertia is difficult to overcome."
He pointed to education technology software, which he said he uses all the time. "It's a nightmare," he said, describing tools that fall short of "how things should have been 10 years ago." He also cited the Internal Revenue Service still using fax machines.
His takeaway was realism. As he put it, "not everything is like startups."
Questions people ask
Is Cowen bearish on AI itself?
No. He says AI benefits will exceed costs, but the path will look more normal than many boosters suggest. In the same conversation, he says AI will likely spread faster than electricity or the internet, while still running into institutional limits.
Why are schools, hospitals, and agencies slower to adopt new technology?
They often have public missions, strict rules, tight budgets, and many stakeholders who must agree before changing tools. Schools and hospitals also handle sensitive data and cannot easily shut down for an upgrade, so older systems persist.
If AI helps people, will that always show up in economic growth?
Measured output can miss free digital tools, better health, and longer lives. Cowen defends gross domestic product, the value of goods and services produced, as useful in the short run, but agrees it does not capture every gain in well-being.

