Developers and AI: A Complex Relationship
Developers won’t work without AI anymore. The research, however, suggests that this reliance might be hindering productivity rather than enhancing it.
The Paradoxical Findings
- METR’s Study: In February 2026, METR attempted to replicate a study on AI’s impact on developer productivity but could not due to developers’ refusal to participate without AI assistance.
- Original Results: Surprisingly, the initial study indicated that AI increased productivity, but subsequent data revealed the contrary—AI slowed down developers due to additional time spent on error fixing and steering AI models.
- Self-Reported Data: A recent survey by METR showed that developers believe AI makes them twice as valuable, but evidence from Amazon and Uber paints a different picture.
Corporate Trends and Challenges
- Tokenmaxxing: Companies like Amazon and Uber have faced challenges with their AI spending, demonstrating that measuring AI adoption solely by token consumption is ineffective.
- Uber’s Budget: Uber blew through its entire 2026 AI budget within four months, indicating a lack of tangible productivity gains despite substantial investment.
- Salesforce’s Projection: Salesforce predicts $300 million in Anthropic token spending for 2026, highlighting the ongoing trend of using tokens as a proxy for productivity.
The Code Quality Concern
“You write code twice as quick now? Better hope you’ve halved your maintenance costs.” — James Shore, Programmer and Author
The issue lies in maintaining code quality. Rapid AI-assisted coding might lead to increased maintenance burdens, ultimately causing long-term problems. As Entelligence AI states, companies spend 44% of their tokens on these less productive tasks.
In Conclusion: While AI offers immense potential, its integration into development workflows should be strategic and focused on enhancing productivity rather than solely measuring token consumption.