A developer who builds AI agent infrastructure for a living is sounding the alarm about AI fatigue - a phenomenon the industry is ignoring. Siddhant Khare, a core maintainer of OpenFGA and creator of multiple AI tools, describes hitting an exhaustion wall despite shipping more code than ever. The paradox is stark: AI genuinely speeds up individual tasks, but the overall cognitive burden has increased dramatically.

The problem isn't that AI doesn't work. It's that AI changes the nature of engineering work from creation to constant evaluation. Before AI, engineers spent deep focus time on single problems. Now, they context-switch between six different tasks daily, each taking an hour with AI assistance. This rapid task-switching is brutally expensive for the human brain, even as managers and engineers themselves adjust expectations upward.

Khare identifies a fundamental shift in job responsibilities. Engineers have become reviewers rather than creators, spending their days judging AI-generated code for correctness, safety, and architectural fit. This evaluative work is psychologically draining in ways that generative work is not. The nondeterministic nature of AI outputs adds another layer of stress, as engineers can never fully trust or predict what they'll get from the same prompt on different days.

Quick Fact: The article notes engineers now context-switch between six different problems daily when using AI tools.