The first bot that tried to replace junior devs (20% success rate)
A March 2023 weekend sprint that auto-generated pull requests from GitHub issues before most people had heard of "AI coding."

What it does
AutoPR is a GitHub Action that watches for an AutoPR label on issues, then plans a fix, writes code, pushes a branch, and opens a pull request. It was built in a weekend right after OpenAI’s first ChatGPT API dropped, making it — by the author’s account — the first autonomous PR bot.
The interesting bit The real curiosity here is historical. AutoPR predated grammar-constrained LLM output, so it used Guardrails to prompt for JSON Schema-structured data and re-prompted when the model hallucinated wrong shapes. The author openly admits it worked about 20% of the time and now frames the repo as “a brief glimpse into the past.”
Key highlights
- Triggered by labeling an issue with
AutoPR - Used Guardrails for structured LLM output before constrained decoding existed
- Generated real PRs: dice bots, weather apps, tokenizer swaps
- Author is transparent about limitations and alpha quality
- 1,372 stars suggest it captured early-2023 imagination
Caveats
- Explicitly archived/retrospective: the author says “enjoy this brief glimpse into the past”
- Real failure modes documented: phantom function calls, duplicated lines, incorrect cross-file references
- GitHub-only; no other platform support
- 20% success rate was considered normal “back in the day”
Verdict Worth a look if you’re studying the evolution of AI dev tools or want to see how far constrained-prompt engineering has come. Not worth installing unless you’re a historian or feeling nostalgic for March 2023.