Geoffrey Huntley discusses the success rate of LLM-driven development agents like Cursor in building projects that meet acceptance criteria. He explores how to utilize Cursor’s rules feature to have it organize and write its own rules. Huntley cleverly made Cursor write a rule to eliminate Bazel as a build system. The article also highlights the benefits of using Semgrep as a semantics-aware code search tool to detect vulnerabilities. The author predicts a shift towards “closed-loop” LLM agent code generation where the system can fix errors and exceptions automatically. This innovative approach could revolutionize code testing and generation processes.
https://fly.io/blog/semgrep-but-for-real-now/