LLMs can be absolutely exhausting

Working with LLMs for extended coding sessions can become exhausting, and the root cause is often the developer's own mental fatigue rather than model degradation. Tired developers write worse prompts, interrupt the AI mid-task, and fall into a doom-loop of poor results. Two key failure modes are identified: degraded prompt quality from fatigue, and slow feedback loops that bloat context windows. The recommended approach is to treat slow feedback loops as a problem to solve directly—by asking the LLM to help create faster, TDD-style iteration cycles with clear success criteria. Recognizing when you're too tired to write a clear, confident prompt is the first signal to take a break rather than push through.