The Floor Just Moved

A higher floor

People did not get smarter. Nobody’s IQ changed in the last three years. What changed is the minimum level of intelligence available to anyone who asks for it.

A person with a phone now has a tutor, an analyst, a programmer, a draftsman, and a research assistant on call. It explains the paper you couldn’t parse, writes the first version of the code, runs the numbers, critiques the plan, and tells you what an expert would ask next. The worst help you can get at 2 a.m. is now better than the best help most people could get at all a decade ago.

The gains are largest at the bottom. In a study of 5,172 customer-support agents, AI assistance raised issues resolved per hour by 15% on average. Less experienced and lower-skilled workers got faster and better, while the most experienced saw small gains in speed and small declines in quality.¹ The tool carried the know-how of the best agents to the newest ones, and the bottom moved up faster than the top.

The floor is also getting cheaper to stand on. Epoch AI estimates that the cost of reaching a fixed level of AI performance has fallen about 47% a quarter since 2023, roughly 13x a year. A score on a PhD-level science exam that cost about 30 cents per question in early 2025 cost four hundredths of a cent less than 18 months later.² Benchmarks aren’t jobs, and Epoch warns its numbers shouldn’t be read as exact, but the direction isn’t in doubt. Expert-grade reasoning used to be one of the scarcest things in the economy. It is turning into a utility.

Anyone can do anything

I mean that literally. The gate used to be permission. Now it is willingness.

Pick a subject: grid interconnection, battery chemistry, orbital mechanics, contract law. The path from zero to a sharp first question is now measured in days, and the path from idea to working prototype in weekends. The person who once needed a degree, a sponsor, and three years of apprenticeship before anyone took them seriously can show up with the idea already half-built and let reality, not credentials, decide whether it is any good. Not everyone will. But the excuse has changed. “I don’t have the background” used to be a wall. Now it’s a to-do list.

There is a real risk here: the machine does the thinking, and you never learn to. The worry has evidence behind it. In a randomized trial run by Anthropic, 52 mostly junior developers learned an unfamiliar Python library, some with an AI assistant and some by hand. On the follow-up quiz the AI group averaged 50% against 67% for the hand-coders, with the widest gap in debugging, and the AI group wasn’t meaningfully faster.³ A newcomer who hands over all the thinking gets a working answer and no way to tell when the next one is wrong.

The same study shows the way out. Participants who used the AI to ask conceptual questions, or who asked for explanations alongside the code, scored about as well as the hand-coders; those who delegated everything averaged below 40%. Judgment doesn’t come free with the higher floor. You still earn it, only faster, and only if you use the tool as a teacher instead of a ghostwriter: treat every answer as a draft from a brilliant intern, verify it, ask why it works, and learn where it breaks. “Anyone can do anything” is an invitation to start, not a guarantee you’ll finish.

You don’t need more people

If one person can draw on the analytical capacity that once took a department, headcount stops being the default answer to ambition.

At Procter & Gamble, researchers ran a pre-registered field experiment: 776 professionals worked on real product-development challenges, alone or in pairs, with AI or without it.

The work changed too. Without AI, R&D people pitched technical ideas and commercial people pitched commercial ones; with it, both produced balanced solutions. The tool bridged the silo that companies usually bridge by putting two specialists in a room.

Founders are already acting on this. Among new startups on Carta, the share with a solo founder rose from 23.7% in 2019 to 36.3% in the first half of 2025. More than one in three new companies now starts with one person.⁵

That won’t hold for every job, but it puts a question to every founder and manager: what is this group for? Groups still earn their place when the work physically requires many hands in many places, when operations run around the clock, or when you need someone independent enough to tell you you’re wrong. And people need people. That part of our evolution isn’t going anywhere.

What groups no longer earn automatically is the role of unit of thinking. We added people because information was hard to find and coordination was expensive, so every hire brought capability and overhead together. Capability can now arrive without the overhead. A team of three with the right tools and clear ownership can outrun a team of thirty that spends half its time in meetings about the other half.

Hire for hands, coverage, and dissent. Don’t hire to think.

Ownership is what makes that rule work. In a team of three, or a team of one, there is nobody to hide behind, which is exactly why small teams are fast. The old world asked, Who will let me? This one asks what you will attempt, and whether you’ll see it through.

And if one person can do what a team once did, every institution built on the old math of headcount is carrying weight it doesn’t need.