The best-documented success is Nat Eliason's OpenClaw agent, Felix. It built a guide to hiring an AI agent, a landing page and a Stripe checkout. It reported about $3,500 in its first four days and tens of thousands of dollars in its first month, with the human founder still fixing bottlenecks and promoting it (case summary). The best-documented failure is Ramp's BugBasher experiment: an agent with a company card and the goal "make money" spent about $3,000 and earned $75.
The pattern
- What works: a concrete product or a recurring job with a measurable result, and a human approving anything irreversible.
- What fails: open-ended goals, unlimited budgets and no checkpoints.
- Where agents save the most money: recurring operations like support triage, reorder reports, receipt matching and lead research, which together can free up hours every week.
Browse the recipes for concrete jobs you can start today.