I’m Erik Hernandez. I build BudgetSimpler, a budgeting app, as a solo developer. AI coding agents write a large share of the code, and the process around them exists to catch what they get wrong before users ever see it.
With Receipts documents that process the only way I trust: through things that actually happened. Real incidents, the CI gates that came out of them, real numbers, and the commit history to back every claim. If a claim cannot be verified against the repo, it does not get published. Where a post makes a specific technical claim, the receipts are linked as gists: diffs, test output, the code as it exists.
The arc, roughly: each post takes one failure that happened in a real production codebase and follows it to the mechanism that now prevents its whole class. Bugs are interesting; the gates they leave behind are the point.
I also do pre-launch audits of AI-assisted codebases. If you are about to put one in front of strangers and want a second set of eyes on it first, reply to any post and it reaches me.
