AI tools like Claude Opus 4.6 make engineers 10x more productive and are addictive, but also drain developers' energy, causing increasingly widespread burnout
This was an unusually hard post to write, because it flies in the face of everything else going on.
Context & Ripple Effects
This report extends coverage of developers describing AI fatigue, a rapid-tool-adoption FOMO cycle and perceived thinking atrophy alongside higher output, as in an engineer's account of the AI-tool FOMO treadmill.
It also sharpens an unresolved measurement problem: a prior METR study found experienced open-source developers took longer with coding assistants despite feeling faster. The reported burnout cost makes the gap between apparent throughput and sustainable performance consequential.
First-order effects
- Engineers using tools such as Claude Opus 4.6 may experience a sharper trade-off between rapid task completion and depleted energy, making sustained use harder even when output rises.
- Engineering managers face a more ambiguous productivity signal: reported gains cannot be treated as a clean proxy for team capacity when intensive use is associated with burnout.
Second-order effects
- AI coding-tool vendors will be pressured to compete on workflow design and sustainable use, not only on headline capability or speed, as fatigue becomes part of the product experience.
- Teams may add more review, pacing and workload controls around agent-assisted work; that can reduce the operational value of apparent speed gains if it is needed to preserve developer capacity.
Third-order effects
- If repeated across organizations, AI adoption in software work will be judged increasingly by durable output per engineer rather than short-run task velocity—a version of the reported AI “brain fry” constraint on cognitive capacity.
- The broader market could separate tools that fit stable workflows from those that intensify constant switching and usage pressure, though the corpus does not establish which products or practices will prevail.
The trend: AI-assisted development is moving from a capability race toward a sustainability test: whether gains in output can be maintained without exhausting the people operating the tools.