Meta's new structured prompting technique makes LLMs significantly better at code review — boosting accuracy to 93% in some cases
Deploying AI agents for repository-scale tasks like bug detection, patch verification, and code review requires overcoming significant technical hurdles.