ironically or not, the best way to have LLMs be effective at writing valid code is when they work on microservices. Since the scope is smaller and the boundary

ironically or not, the best way to have LLMs be effective at writing valid code is when they work on microservices. Since the scope is smaller and the boundary is clear, tools like Cursor/Windsurf seem to make very few mistakes (compared to pointing them at your monorepo, where they usually end up completely wrong)