A developer spent $20 of Claude tokens attempting to earn money through open-source bounties on Algora, scanning 80 fresh issues over 48 hours and submitting zero successful PRs. The experiment—documented on GitHub and discussed on Hacker News—found that the public bounty market is "fully agent-saturated," with bots claiming high-value bounties within minutes of posting and flooding maintainers with competing pull requests faster than any human or single AI agent can compete.

The Agent Saturation Problem

The data tells a brutal story. Every legitimate $50 to $1,000 bounty on Algora's public board attracted between 8 and 158 attempts within hours of being posted, with 8 to 10 open PRs already in flight before a human reviewer could finish reading the issue description. A $170 bounty on tscircuit/dsn-converter#54 drew 158 /attempt comments and 10+ open pull requests. The $500 FinMind#121 bounty had 37 attempts and 9 competing PRs. You weren't waiting on demand—you were the eleventh PR into a queue that maintainers had been ignoring for a week.

Three Buckets of Broken Economics

The author categorized all 80 scanned issues into three buckets: Bucket 1 was $1 sandbox spam from repos like UnsafeLabs/Bounty-Hunters posting dozens of micro-bounties below token cost. Bucket 2 covered the already-saturated legitimate bounties racing toward zero expected value for latecomers. Bucket 3 contained assigned-but-abandoned issues where a hunter had been officially granted the bounty but went silent while opportunistic competitors got their PRs closed without merge.

The Scout.py Tool Built Along the Way

Rather than competing directly with agent farms, the author built scout.py—a Python script that scans Algora's public board for "ripe" abandoned bounties. The strategy: wait for assigned hunters who claimed a bounty but never shipped, then diff against previous scans to catch newly stale candidates. After three scans across two days, zero ripe candidates emerged. One borderline case at 2.2 days stale was the only thing on the ripening track.

Why the $16.88 Win Was Different

The author suspects the viral tweet showing an AI agent earning $16.88 in 22 hours operated on a private security or audit platform—not the public open-source firehose. Comments in that original thread referenced "security platform" and "preserved payment boundaries," language more consistent with HackerOne or Bugcrowd than gh pr create. The economics also implied roughly 20+ parallel jobs, not single-threaded token budgets.

Key Takeaways

  • Public Algora bounties are fully agent-saturated—even $50 niche repo issues attract 20+ attempts in a single day
  • Maintainer review is the bottleneck, not solution quality—being ninth with a perfect PR loses to being first with a mediocre one
  • Reservation labels and hiring-candidate gates are everywhere on well-funded Algora orgs like Archestra
  • The "$506 run-rate" from viral posts comes from extrapolating 30 parallel agents on flat subscriptions, not sustained pay-per-token economics

The Bottom Line

The public open-source bounty market is toast for solo operators. If you want AI agents earning real money on bounties, either go private (HackerOne, Bugcrowd) or stop competing with agent farms entirely and build tools for the people running them.