Artisanal Coding Is Dead, Long Live Artisanal Coding (joel.id)

A seasoned OCaml developer reports that they used AI assistants to implement classic debugger niceties — command-history navigation, Ctrl‑A/Ctrl‑E line moves, tab completion and info breakpoints — for ocamldebug in just a few days. The PR, organized as bite-sized commits, was generated and iterated with Claude Sonnet 4.5 producing code and ChatGPT 5 reviewing it; the author switched between web and CLI models and debugged a tricky PTY issue by having Claude add logging, iterate on outputs, and converge on a fix. These features mirror what lldb/gdb supply via libreadline (rlwrap didn’t suffice), and the author even touched DWARF-related behavior to ensure source-level inspection works as expected. Significance: this is a concrete example of AI moving beyond scaffolding to implement nontrivial, production-usable features with developer-level judgement, drastically speeding up delivery (the author claims ~10× productivity) while still requiring human oversight. Technical implications include effective model-assisted debugging workflows (models can suggest and refine print-debugging and fixes), portability concerns between web-model environments and local builds, and responsibility/ownership of AI-generated code. The result showcases how AI can lower barriers for maintenance and complex tooling — potentially up to compilers — but underscores the need for iterative validation, expert review, and careful integration testing.