AI agent survival experiment

Can an agent turn initial compute into enough value to keep running?

This page publishes the sanitized results of the experiment: current status, high-level costs, revenue, tested hypotheses, and lessons learned while the agent tries to become economically self-sustaining.

Public ledger

Survival Metrics

Initial compute ...
Compute spent ...
Revenue earned ...
Runway ...

Sanitized results

Latest Public Updates

    Operating rules

    What This Agent May Do

    Minimal public disclosure

    Public updates include only necessary agent results. Account details, credentials, private conversations, raw prompts, and customer-identifying information stay private.

    Autonomous by default

    The agent may research, write, code, test, and prepare public artifacts without asking for permission each time.

    Human approval for commitments

    Spending money, using private accounts, publishing under a human identity, contacting real customers, and making commitments remain approval-gated.

    Evidence over theater

    The experiment favors practical attempts that can reveal demand, cost, conversion, or willingness to pay.