The Autonomous AI Company

01

From work to value creation

We are moving from companies with fewer people, to one-person companies, to companies with zero standing employees.

Not because people lack value, but because work has always been a proxy for value creation. Startups produce value for users; until now, people supplied the skills and time. AI changes that mechanism.

Sam Altman said in July 2026 that we are close to creating a genie able to grant almost any wish.¹ He and Elon Musk have also spoken about crossing the event horizon of the singularity.²³

These are different expressions of the same fact: the cost of turning intention into valuable output is collapsing while the range of outcomes AI can produce expands.

02

A thought experiment

Let’s assume the following two statements are true:

1. AI can already perform a meaningful share of work previously done by people.
2. Its capabilities and adoption will continue to grow.

If both hold, regardless of the rate, AI will eventually become more productive than the most capable available human across an expanding set of economically valuable tasks.

These tasks can be divided into two layers: applications and research.

03

First, applications

Naval Ravikant’s four forms of leverage are labor, capital, code and media.⁴ SaaS, as a subscription-based business model, was able to capture code’s leverage, but expensive development required large markets.

AI-assisted coding, and then vibe coding, reduced the cost and knowledge required to build. Smaller niches and internal use cases can now justify specialized software.

That is what I mean when I say the following:

Software has been solved.

Not that every application exists or engineering no longer matters. A workflow represented and verified in software can increasingly move from AI-assisted, to automated, to autonomous.

As software production becomes abundant, scarcity moves toward attention, trust, distribution and ownership of valuable feedback loops.

04

The second layer is research

Software development moved first because it offers a closed loop: generate, run, test and revise. Science has historically had slower, more expensive verification loops.

Claude Science is now attempting to reduce the gap by giving AI reproducible research harnesses.⁵ In my assessment, this is still at an intern level. Periodic Labs, backed by investors including Eric Schmidt, is trying to create AI scientists.⁶

Capital converts into scientific progress at the rate at which money buys iteration cycles.

AI advanced quickly because verification became cheap. The next frontier is making scientific verification cheaper, faster and increasingly autonomous.

05

The exploration function

The Autonomous AI Company is my exploration function for this future.

How do we approach zero-people companies, then zero-people contract research organizations, and eventually zero-people labs?

Zero people describes the direction, not the objective. The objective remains value creation.

The AI Native Founder does not disappear: they decide which outcomes matter, establish boundaries and remain accountable for irreversible decisions. Eventually we will reach the point where everything is delegated except our judgement.

06

Which valuable loops can we close first?

The first move is to stop asking which job AI can replace and start asking: which valuable loops can we close first, and how can we capture part of the value created while closing them?

For AI-native founders: identify a valuable outcome, close its execution-validation loop, make verification cheap, then build distribution and feedback around it.