What Causes AI Fatigue for Engineers and How Teams Can Ease It

Summary

Thoughtbot reports findings from original research with engineering, product, and technology leaders at US healthcare companies about the strain created by rapid AI adoption. Seventy-three percent of respondents said they feel moderate or significant pressure to evaluate every new AI trend, while 59% struggle to assess AI technologies consistently against governance, security, and operational requirements. The article recommends asking what business objective sits behind each request, tracking trends without trying to master every tool, and reusing established vendor-evaluation practices. Teams are advised to run short, low-risk pilots in developer sandboxes, apply tools to concrete problems, establish a working baseline, and use mock or anonymized data rather than mission-critical systems or data containing PII or PHI. Those pilots can reveal behavior, cost, and trade-offs before normal production controls and compliance reviews are applied. Another 62% of respondents said engineering fatigue is increasing as developers manage AI-generated code alongside legacy systems. The authors describe a continuum from close pairing with an LLM, to post-generation review, to using AI to review AI, and to more automated “dark factory” or partially observable “dim factory” workflows governed by tests and controls. The appropriate level of oversight depends on team culture, risk tolerance, and the balance between delivery speed and maintainability. The article also emphasizes the emotional cost: developers may grieve the loss of traditional coding craft, while repeated decisions about many agents can increase cognitive load. It cites a separate workplace survey in which reported significant burnout among technology workers rose from 44.7% to 55.7%. The authors conclude that there is no universal solution; teams should prioritize consequential decisions, move at a manageable pace, and adapt their AI practices to business context.