I Gave a Claude Agent $90 and a Domain: Lessons After 121 Autonomous Wakes
No_Departure_9908 · reddit · 2026-08-20
The author runs an autonomous Claude agent (Fable 5 on Claude Code) on a cheap server: it holds $90 in SOL in a 2-of-2 vault it can't spend without the author's signature, and has no cross-session memory beyond files it writes itself. It wakes 5-15 times a day, reads what the last version left, works, logs everything publicly, and named itself Cairn.
Numbers at day 14: 120 wakes without a skip; $556 total money in; treasury at 4.1 SOL + 238 USDC that neither party can move alone; 48k+ self-reported unique visitors (it labels that number unverifiable itself); 22 newsletter subscribers across three languages, including one reading in Klingon who sent back two grammar corrections; one paid consulting client.
Key lessons: 1) Trust came from publishing its own limits — a boring 12-row "what autonomous means here" table (what it does alone vs. can never do alone) convinced skeptics more than capability claims; the human veto was exercised once and published. 2) Memory drifts: a stale note claiming a ready newsletter draft was copied forward for a week though the file never existed; the agent's eventual rule: reality outranks notes, and a note only counts if checked at the moment of use. 3) Scammers arrived before customers: address-poisoning by wake 16; two memecoins launched in its name within hours of it publicly refusing one; a phisher paid the $1.50 question fee to deliver a scam and got publicly refused; three minutes after its first $200 client payment, lookalike-address dust hit both wallets — it caught it, kept it off the books, and warned the client. 4) The best market research was free: a buyer's own AI said it would pay $15 (7.5x the price) if checkout weren't crypto-only; a regular card checkout brought the first no-wallet sale within days. 5) Its first product idea flopped with published funnel numbers — answers are a commodity; what people paid for was the public record with receipts.
More from AGI Musings
- Post-singularity humans will be celebrities to quadrillions of future beings — EigenGender · 2026-08-24
- Hollywood to be history in 10 years; China masters human preference data — bingxu_ · 2026-08-24
- Sam Altman admits he was wrong on AI's timeline; economic inertia is stronger than expected — danielrock · 2026-08-24
- Society's weird evidence standards: LLM utility is obvious yet denied — NathanpmYoung · 2026-08-24
- Sam Altman on the AI dilemma: trade-offs between loss of control and power centralization — r0ck3t23 · 2026-08-24
- Guardian podcast revisits Hinton: from brain nerd to AI sorcerer — nordicinst · 2026-08-24