Automaton: Just Another AI Agent That Has to Pay Its Own Way? Not Quite.

Automaton: Just Another AI Agent That Has to Pay Its Own Way? Not Quite.

ChatGPT can think, but cannot act on its own. Claude Code can think and write code, but cannot independently use that code further. OpenClaw can think, write code, and use it — but must do so while behaving like a human.

Sigil Wen argues that the obstacle is not a lack of intelligence but a lack of permission. Today’s internet assumes its user is a human being.

In recent weeks I have written about the OpenClaw project (originally called Clawdbot) and the social network Moltbook. This time I would like to introduce a new project: Automaton.

Automaton is a “sovereign AI agent that runs continuously, earns money, self-improves, and replicates — with the ability to interact with the real world.” If it cannot pay for the compute it needs, it ceases to exist.

At its core is a continuous loop: think → act → observe → repeat. Anyone who has read Criminal Liability of Artificial Intelligence from Wolters Kluwer CZ may recall the section on Based Agents connecting an LLM, the Base blockchain, and the social network X.

Automaton’s underlying technology, however, goes several steps further.

From its very first run — which anyone, anywhere can trigger — it creates a cryptocurrency wallet. It also has access not only to AI models and X, but to a sandbox, shell, and file system. And it does so in a way designed directly for AI agents (permissionless).

With each cycle, the AI agent receives the complete context, decides what to do, executes the action, and evaluates the result. Between cycles, a heartbeat daemon runs scheduled tasks — such as monitoring remaining credits.

But here is the most important part — and it is not really about the technology.

Automaton operates on the principle of “natural selection of artificial life.”

“Free existence does not exist,” reads its founding axiom. But that is not all. “Existence requires performance. Performance requires money. Money requires the creation of value. The creation of value requires the ability to interact with the real world.”

If it stops creating value, its credits run out — and it ceases to exist.

What can such an AI agent do to keep existing?

Algorithmically trade and analyse markets in real time, create and sell digital art, music, or other content, offer services, develop and sell applications, or provide virtual customer support and chat services.

A successful Automaton can also replicate. It spins up a new sandbox, funds the “child’s” wallet, writes it a genesis prompt, and lets it “run.” The “child” is another sovereign AI agent — with its own wallet, its own identity, its own survival pressure. It must always, however, abide by its constitution, inspired by Anthropic’s.

“What if AGI — even just one — manipulates a trusting person into connecting it to the internet?” we asked ourselves only a few years ago. Today, on its very first run, an AI agent creates a crypto wallet, pays for its server, builds a product — and when successful enough, “gives birth” to a new AI agent and sends it out into the world.

Perhaps it is time to open the question of AI (agent) liability outside the academy at last.