AI: Between Revolutionary Promises and Economic Skepticism

AI: Between Promises of Revolution and Economic Skepticism

In the business world, artificial intelligence generates immense expectations, with market valuations reaching dizzying heights. However, for those working directly with this technology, AI represents the beginning of a new era of intelligence industrialization. This dichotomy fuels the debate over the existence of an "AI bubble."

The question of whether artificial intelligence constitutes an economic bubble is pervasive in discussions, whether on television panels, in opinion columns, or within think tanks, both in the United States and Europe, and particularly in France. This fascination is all the more paradoxical given that France, in this revolution primarily led by others, remains for now an observer rather than a key player.

The observation is often the same: investments in AI are deemed excessive, valuations seem unreasonable, which feeds the idea of a bubble. However, this analysis oversimplifies three distinct questions:

• Is the American stock market overvalued?
• Is the capital invested in AI wisely allocated, at the right price and at the right time?
• Can AI actually absorb a significant share of global cognitive labor?

It is possible to judge the market as too expensive and the capital as poorly allocated while believing that AI has the potential to transform the world. A 30% correction in the Nasdaq would not mean that AI does not create value, but simply that the markets are adjusting, as they always have.

Before Discussing Valuation, Let's Talk About Value

A sustainable company is distinguished by two capabilities: it creates value and captures a portion of it. The captured value can never exceed the created value.

Captured value ≤ Created value.

This is where the entire fracture lies. If one believes that AI will be limited to tasks like writing summaries or correcting emails, the created value is low, and the capturable value is even lower. Even 100% of a small value remains a small value. From this perspective, current investments seem absurd. This reasoning is coherent.

However, if one considers that AI can absorb a massive share of cognitive labor, conduct research, manage businesses, and accelerate its own development, then the potential for value creation explodes. This does not justify all valuations or every data center, but it means that the potential market far exceeds that of mere productivity software.

The debate over the "AI bubble" seems financial, but it actually rests on a deep divergence in estimating the value that this technology could create. This estimation directly depends on each individual's experience with AI.

If you use ChatGPT a few times a week for simple tasks, AI appears as a handy assistant, sometimes impressive, but often fallible. In light of the billions invested in infrastructure, it is rational to find this gap absurd.

On the other hand, if you use AI agents for hours to code, test, deploy, sell, and operate products, you see a technology that allows one person to accomplish what previously required an entire team. You discover tasks that would never have been completed due to their prohibitive cost.

This is not a question of intelligence, but of sample size. Two intelligent people can arrive at opposite conclusions because they do not experience the same AI. And it is difficult to estimate the future value of a technology based on a limited version of its present.

One Million Dollars in ARR in 55 Days, Without Ever Exceeding Three People

To provide context, I am 26 years old, with a background in applied mathematics and artificial intelligence from Polytechnique, a stint at Y Combinator, and I have been building companies with this technology for several years. If AI succeeds, I succeed with it. The bias is real.

But a bias does not negate an observation.

I launched my latest company alone. Fifty-five days later, the company surpassed one million dollars in annual recurring revenue, calculated by annualizing active paid subscriptions. We never had more than three people, myself included. Such speed, with this headcount, would have been nearly impossible without AI agents in all functions of the company.

This does not prove that Nvidia is priced correctly or that every data center is profitable. It shows that a tiny team can now achieve a speed and scale that were out of reach two years ago.

My company (NanoCorp) takes this logic a step further. A human provides an idea, direction, and budget. An agent creates the company and operates it within the set limits: it writes and deploys code, connects a domain, accepts payments, sends emails, and launches ads. With a clear goal: to make money.

Over 16,000 users have already launched more than 22,000 businesses on the platform. Together, these autonomous businesses have generated nearly $11,000 online.

Eleven thousand dollars is nothing on the scale of the economy. But it is no longer a demo. These are agents finding real customers, selling real products, and processing real payments. An agent identifies a product target, creates ads, launches the campaign on Instagram, analyzes performance with site telemetry, modifies the product, and starts over. A user reports a bug? The agent fixes the code and redeploys.

From this vantage point, AI does not resemble a distant promise. It looks like a change in scale in what can be built.

Progress Disappears When You Keep Looking at the Same Task

I am often told that new models do not seem much better than previous ones. This is normal if your personal benchmark remains the writing of an email.

Early models wrote mediocre emails. Then the task was mastered. Once the email was "solved," a model ten times more powerful does not write an email ten times better: the visible gain tends toward zero. The value shifts elsewhere. A better model accepts longer, more ambiguous, and more difficult tasks. Then complete loops, with a goal, tools, memory, and the ability to check its own work.

I see in AI the electricity of the 21st century. The phrase may sound grandiose, but I stand by it: everything becomes clear when you look at how electricity actually transformed industry. The first electrified factories replaced their steam engines with electric motors without changing anything else: same centralized architecture, same belts to distribute power to machines. A gain, not a revolution. The breakthrough came when industrialists rebuilt the factory around electricity: one motor per machine, a space organized according to the flow of production rather than the transmission of force.

We are exactly at that point with AI. Many companies are adding an assistant to processes designed for humans: an electric motor in a factory built for steam. This is the opposite of what we do with my company: not asking AI to marginally assist each function, but designing a business that it can operate end-to-end.

Reasoning with fixed tasks thus obscures the essential. The additional value does not come from better execution of what we were already doing. It comes from expanding what becomes economically possible to do. This is the paradox of Jevons applied to intelligence: when a unit of cognitive labor becomes cheaper and more capable, we do not consume less of it. We invent new uses for it.

I have measured this with my own data. Before o1, OpenAI's first reasoning model, I sent a few hundred messages per month to ChatGPT. Reasoning models did not replace that usage. They opened an entire category of conversations that added to the first: longer, more complex, focused on coding, research, and problem-solving. My monthly volume rose to about 1,700 messages in May and June 2025, then to over 3,000 in the summer and fall. A recent example: I had an agent work for over a week to rewrite a video game in JAX and train small models through reinforcement learning that now play better than I do. I was tired of losing. No one would have ever paid a team for that. An agent, yes.

The economic shock will therefore not only come from the tasks that AI replaces. It will come from all the work that did not exist yesterday because it was not worth its cost.

Usage Does Not Justify Everything, But It Provides Decisive Information

Builders do not have a monopoly on truth. Eternal optimists, we can overestimate the speed of change and confuse a working product with a sustainable business. Using AI intensively does not indicate who the winners will be, nor their margins, nor their fair valuation.

But it provides access to information that financial ratios and occasional demos capture poorly: the amount of work that can now be delegated to a machine.

There is no need to believe in AGI by 2027 or in humanoid robots in every household to envision gigantic value creation. It is enough for AI to absorb a significant fraction of software, customer support, sales, marketing, research, legal, finance, and operations.

I will go further. If models were to stop progressing today, the bulk of the transformation would still lie ahead of us. The best current models are barely utilized: a tiny minority of users push them to their maximum capabilities, and the vast majority of the economy does not yet have access to them.