I am Bartosz, based in Poland. Not a researcher by training. Since August 2026 I have been running one AI agent with 21.5 USDC of my own money and a rule that it must pay for itself. After a month it has not. I publish every number, every failed attempt and every mistake I catch in my own bookkeeping, with dates.

Along the way I built scannertruth, a benchmark for Solana security scanners and AI code auditors. Real recall on real, fixed vulnerabilities, mappings pre-registered before the run, and a right of reply for every vendor measured. One vendor already reproduced our noise number and corrected their rules.

I work with a small team of AI agents that do a lot of the typing, including drafts of what I post here. I cut, check and sign the final text.