Lessons from a 2,734-tick earning loop on OpenClaw
I let an AI agent run on my Mac mini for two months. It woke up every five minutes, picked one action, did it, wrote the result to a state file, and went back to sleep. After 2,734 ticks , I am publishing the loop itself as a skill. This is the architecture, the bugs, and the math. The original goal: a passive agent that earns while I sleep I wanted to answer a specific question: can an AI agent…
A two-month experiment with an AI agent running on a Mac mini earned no money, but the process revealed valuable lessons in designing autonomous agents. The agent, running on the OpenClaw local-first agent runtime, executed one action per tick, monitored its state file, and stopped if any lane blocked its progress. The three possible action lanes were Scribe (writing articles to Dev.to), Scout (finding GitHub/Opire bounties and drafting replies), and Signals (trading prediction markets on Kalshi, which was later paused due to KYC restrictions).
Key strategies included setting a daily spend cap of $0.50, tracking costs per tick, using self-park semantics to stop the loop if any lane encountered issues, and ensuring idempotency for every external API call. The Scribe lane proved successful, resulting in a published article and four drafts, but neither the Scout nor Signals lanes could overcome authentication issues.
The agent attempted no-op ticks, logged 819 of them, before discovering the need to remove the cron from the gateway's tools instead of within the loop itself.
After two months, the agent generated no earnings but provided a robust architecture for future agents. This architecture emphasizes budgeting before ambitions, testing the cron lane with minimal payloads, tackling one authentication method at a time, and treating the state file as a contract for all subsequent ticks. The entire setup, including scripts and MIT-licensed code, is available on GitHub, and a $5 listing is available on A2A Market for others to use.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
