A multi-agent platform that turns a prompt into generated code, tests, a design document, and an audit trail — built for regulated teams starting new projects under contract. This is the debugging journal: append-only, honest about what broke, and specific about why.
Two problems six weeks apart that turned out to be the same problem. A cancel button that stopped the UI but not the work. A search that confidently returned a table-of-contents fragment. Both fixed by refusing to make the primary mechanism smarter.
Read more →Two runs of the same prompt produced byte-identical test files — both containing a test that asserted a plus sign would lowercase to False. The system wasn't broken randomly. It was broken reproducibly, which is worse. Plus the architectural decision that removes a whole risk class.
Read more →Codegen wrote source and the tester independently guessed what was in it. The reviewer judged code it couldn't see. A fallback string became a project requirement and four stages built on top of it. Every implicit assumption between agents was a place a bug could hide.
Read more →I spent days convinced the model was the problem. Random hangs, empty error messages, runs dying mid-pipeline. It was a file watcher, a timeout, and an event loop policy. When part of your system is non-deterministic, every other bug gets to hide behind it.
Read more →Where it started. The architecture, the agent roster, the war stories. What it takes to build an agent platform for the software development lifecycle — and what I'd do differently.
Read more →