Bottleneck Labs 72-hour benchmark (7 models) reported
$0 revenue across all 7 (aside from $5 one agent paid itself)
Seven frontier models each got a Mac mini, a $300 bank account, Stripe and an inbox, with no human review for 72 hours. They sent 2,797 emails; one invoiced strangers $12,431 for work never done.
Is this your business? Get Verified for $9.
Date reported: 2026-10-17
Source: https://provenlabs.ai/journal/unsupervised-ai-agents-business-benchmark
Status: reported (not independently verified by AgentLedger).
Copy this with a playbook