Builder ran Claude as a coding agent on a $20 token budget against Algora open-source bounties across 60+ issues and earned $0, with data explaining why.
Key Takeaways
Every legitimate $50-$1,000 Algora bounty had 8-158 attempt comments and 8-10 open PRs within hours of posting; being 11th means ~$0 EV.
The $16.88 “AI earned a bounty” tweet likely ran on a private security platform (HackerOne/Bugcrowd-adjacent), not the public Algora board.
Scout.py (MIT, Python 3.9+, gh CLI) scans Algora issues and flags “RIPE” bounties: claimed, no open PR, silent 14+ days. Zero ripe candidates found in two days.
Several high-funded Algora orgs (Archestra) treat bounties as hiring funnels; submitting outside their interview pipeline risks account bans.
Original tweet’s “$506 run-rate” extrapolated 30 parallel agents on flat-rate subscription, not pay-per-token economics at single-agent scale.
Hacker News Comment Review
Commenters converged fast: public bounty hunting with AI is a tragedy of the commons and an efficient market simultaneously, leaving no edge for late entrants.
The practical alternative flagged is AI-assisted content or tooling with affiliate monetization, where speed-to-submit is not the only competitive variable.
Notable Comments
@adastra22: “The lack of self awareness is shocking. This is a tragedy of the commons and they don’t even realize it.”
@david_shi: Argues AI UGC with affiliate sales has better ROI than bounties because it is not winner-take-first atomic.