Simple LLM Habits Most Developers Miss, According to a Seasoned Engineer

A developer writing on DEV Community argues that most people adopted poor habits when large language models quietly entered their workflows without any formal training or guidance. The author's core advice is to always prompt the model for a second pass — asking it to identify flaws or gaps in its own output — before treating any result as final. While this self-review trick catches surface-level errors cheaply, the author stresses it does not replace human judgment, particularly when choosing between multiple valid solutions under real-world constraints. The piece also pushes back on debates over LLM code style, arguing cosmetic preferences should be automated via linters rather than debated by engineers. The central warning is that outsourcing one's own reasoning to the model entirely inverts the relationship — making the tool the decision-maker rather than the developer.