Ralph Wiggum loop

A technique for agentic codingagentic coding, n.: Writing software by directing AI agents that plan, edit and test code themselves, with the human acting more as a reviewer and manager than as the person typing every line. in which the same prompt file is fed to a coding agentcoding agent, n.: An AI agent that writes and fixes software by itself: it reads a codebase, edits files, runs commands and tests, and keeps working until a feature is built or a bug is fixed. repeatedly in an unattended loop (at its most minimal, one shell command that keeps piping the prompt to the agent) so that each iteration starts with a fresh context windowcontext window, n.: How much an AI model can read and keep in view at once: the instructions, the conversation so far and any attached documents. Text that does not fit has to be cut, summarized or left out., completes one item from a plan and records progress in files that the next iteration reads. Tests, type checkers and static analyzers act as 'back pressure' that rejects bad output, and expensive work such as summarizing test results is delegated to subagentssubagent, n.: A helper AI agent that a main agent starts to handle one piece of a job; it works in its own separate workspace and reports back a short summary when it is done.. Its proponent presented it as deliberately naive but effective for greenfield projects — citing work worth a sizable contract completed for a few hundred dollars of model usage — while warning against using it on existing codebases.

I left a Ralph Wiggum loop running on the parser overnight; by morning it had passed 900 of 1,000 tests.