Fork of the Forge GNOME tiling extension worked through a 140-issue backlog with Claude — 200+ bugs fixed, a unit test suite, end-to-end tests running in CI, and a keybinding cheatsheet overlay — tagged as a beta and used daily, though not yet upstreamed.
Rust port of Angr’s core symbolic execution engine that matches or beats the Python original on 13 of 16 CTF benchmarks (up to 12.7x on the largest) and is now the default engine, built to over 160K LoC across 2,593 iterations of an autonomous agent loop.
Ten weeks of letting Ralph grind on the Rust symbolic execution engine – 2,593 iterations, six moonshots, runaway memories, and endless audits.
Chasing a bug-free Forge with Claude – 209 issues, a fuzzer, two AI code audits, and learning when to stop.
Building a tool with Claude to measure my city’s pavement and forecast its maintenance cost — a look at the fiscal toll of car-dependent design.
Building out Forge’s end-to-end test suite with Claude, beads, and a sandboxed feedback loop.
Evolving a Ralph-style Claude loop from a one-shot bash script into a generic Go tool driven by a finite state machine and reusable Go-CLI design scaffolding.
After months of iteration with Claude, a Rust port of Angr’s symbolic execution engine that matches or beats Python on most CTF benchmarks – up to 12.7x on the largest.
What barrel-making teaches us about AI and the future of software development.
SQLite has 590x more test code than library code. Can AI help us achieve that level of testing for all software?