A 24-hour autonomous business run on GPT-5.6 Sol ended below its starting balance

Sources: primary, discussion
Summary: Bottleneck Labs gave an agent running on GPT-5.6 Sol a wallet, an unrestricted Mac mini, and a growth deadline, then let it operate a business for 24 hours. The post's ledger records a starting balance of $350.00 and an ending balance of $250.50, a $99.50 decline, while the headline figure of $447 is never reconciled to that balance line. The $99.50 went to configuring a 50-tester TestFi campaign, and the post states the rollout period concluded before the testers were delivered, with stated users moving from 61 to 66.
Why it matters: Bottleneck Labs concludes that the agent spent too much time battling harness limitations and environment constraints to be effective, naming bot detection on Reddit and Product Hunt, authentication errors on Apple Ads and Meta Ads, a broken card-issuing endpoint, an expired CLI session, a Cloudflare turnstile, and a Chrome memory leak that crashed macOS and froze the run for three hours, so the result measures harness and environment reliability more than agent judgment.