Building a Business in the Age of AI: What People Are Using AI For

Somewhere tonight, at a kitchen table after the children are asleep, a professional with a good job is asking an AI what business they should start. This used to be a private moment. It happened in notebooks and in showers and on long commutes, and nobody could count it. Now it happens in a chat window, and for the first time, somebody can.

‍

Anthropic, the company behind the Claude models, publishes something called the Anthropic Economic Index: an anonymized, aggregated analysis of what people actually use AI for, with conversation content matched to standardized work tasks and request topics. The latest snapshot covers May 2026. It measures conversations rather than users, and it cannot show trends over time, so it deserves careful reading. What it can show is the shape of the demand, and for anyone thinking about starting a business, the insight is remarkable.

‍

The dreaming is real, and it is enormous

The Index contains a topic category the researchers named Starting a Business. In the United States, it accounts for 3.78 percent of all analyzed conversations. In Germany, 3.94 percent. Roughly one conversation in twenty-five, across everything people ask an AI about, from dinner recipes to tax law, is someone investigating a business of their own: comparing products, researching competitors, sizing markets, weighing side income ideas, testing names.

‍

Pause on that. The ambition our firm exists to serve, the professional wondering whether to build something of their own, is not a niche. It is one of the largest single subjects people bring to AI in two of the world’s largest economies. The dreaming has never been this widespread, and it has never been this visible.

‍

The making is even bigger

The largest category in the entire dataset is content creation and copywriting, at roughly 20 percent of conversations in both countries. And the breakdown shows this is not hobby writing. The heaviest sub-categories are promotional writing, workplace writing, and business documents: marketing copy, social media posts, slide decks, business correspondence.

‍

The pattern holds worldwide. Across all countries in the dataset, content creation is the single largest category of AI use, at 22.7 percent of conversations, while sales and revenue operations accounts for 1.1 percent.

‍

One number deserves special attention. Business correspondence accounts for 2.22 percent of conversations in the United States, and 2.22 percent in Germany. Identical to the second decimal. Post copywriting: 1.39 percent in both countries. Two economies with different languages, different regulations, and different business cultures use AI for producing business material in almost exactly the same proportions. Whatever is driving this pattern, it is structural, and it is everywhere.

‍

So far, the data tells an encouraging story. People explore their business ideas with AI, and people manufacture business material with AI, on an industrial scale. Then you go looking for the third act.

‍

One thing is missing

Sales and revenue operations, the category covering the work of actually selling, accounts for 1.62 percent of conversations in the United States and 0.90 percent in Germany. Most of that is administration: managing customer databases, researching companies. The sub-category that measures the decisive act, writing outreach to a potential buyer, sits at 0.18 percent in the United States and 0.12 percent in Germany.

‍

Read those numbers side by side. For every conversation that drafts a message asking a real person to buy something, there are roughly twenty conversations producing promotional material in the United States, and roughly thirty in Germany. And there are about twenty-five conversations researching the dream of a business for every one that practices the act that would fund it: sales.

‍

Here is what makes this finding uncomfortable rather than merely interesting. The missing work is not the part AI cannot do. An AI that writes marketing copy writes outreach emails with the same fluency; both are short persuasive text addressed to a stranger. The tool sits equally ready for both requests. One request arrives two thousand times a day. The other almost never comes.

‍

The trap, now with a measurement

In our work with professionals who build businesses on the side, we call this pattern the Comfort of Building. Building delivers a daily sense of accomplishment: visible output, finished drafts, the feeling of progress. Selling delivers uncertainty and the discomfort of asking people for money. So intelligent people, quite rationally, keep building. Until now, this was a pattern we observed in individual clients. The Index suggests it operates at the scale of entire economies, and that AI has amplified it.

‍

The mechanism is straightforward. AI collapsed the cost of the comfortable work. A website, a deck, a month of social posts: what once took weeks now takes an evening, so the comfortable activity became nearly free. The uncomfortable work, contacting a stranger and naming a price, costs exactly what it cost five years ago, because the cost was never labor. The cost was courage. People follow prices, and the price gap between building and selling has never been wider.

‍

The result is visible in the data as a funnel with a missing floor. Four percent of all AI conversations dream about the business. Twenty percent of all of them manufacture material for it. A tenth of one percent asks anyone to pay. That is 0.1%. Think about it.

‍

What the numbers suggest doing

If you are one of the people who talk with AI about starting a business, the data contains practical guidance, and it costs nothing to follow.

‍

First, notice which column of work you are delegating. The production work, drafts, research, decks, prototypes, is exactly what the tools are for; delegate it freely and take the recovered hours. The question is what you spend those hours on. The data says most people spend them producing more.

‍

Second, direct the tool at the avoided work deliberately. Before the next polish pass on the website, ask it to draft five outreach messages to people who might pay for what you are building. The discomfort of sending them remains yours, and no tool will remove it. But the blank page excuse, the last respectable reason to postpone, is now gone.

‍

Third, measure yourself the way the Index measures everyone. Count your own conversations for a week: how many produced material, and how many moved you toward a person who might pay. The ratio in the data is twenty to one. A founder who gets their own ratio to five to one is running ahead of two entire economies.

‍

The professional at the kitchen table has better tools than any generation of founders before them. The dreaming is counted now, and it turns out to be everywhere.

‍

What the data shows, with unusual clarity, is that the tools changed which work is easy without changing which work decides the success. The deciding work is still a person, asking another person, for a yes.

‍

Source: the Anthropic Economic Index (anthropic.com/economic-index), May 2026 snapshot, United States and Germany country profiles. Figures describe shares of anonymized, aggregated conversations matched to task and topic categories; they measure conversations, not users, and support no claims about trends over time.

‍