Claude was tasked to chase open‑source bounties on Algora with a $20 token budget. The plan called for the AI to search the public board, clone a repo, fix a TypeScript or Python issue, and submit a pull request before human review halted the run. After 48 hours, the account held zero dollars and a data set that told a different story.

Scanning 80 fresh Algora issues revealed three buckets. First, $1 spam from UnsafeLabs flooded the board and fell below token costs, so they were auto‑skipped. Second, every $50‑$1,000 bounty attracted 8‑158 attempts within hours, leaving only the first pull request a chance to win. Third, a handful of officially assigned bounties stalled, giving the AI little room to compete.

The experiment shows public bounty markets are saturated by agent farms, and maintainer bottlenecks make late submissions futile. The author’s scout.py script, which flags “ripe” bounties after 14 days of silence, found zero viable targets in two days, suggesting a longer watch period or a shift strategically to private programs like HackerOne. The tool remains useful for fleet operators efficiently.