Developers refuse to work without AI as research flags potential pitfalls
The Story
The TechCrunch article reports that in February 2026, AI research lab METR found most developers are unwilling to work without AI, even for limited tasks. This disrupted METR’s plans to repeat a 2025 study that had shown AI-generated code actually slowed developers down. A May 2026 survey allowed self-reported productivity gains, with developers perceiving AI doubled their value, but recent trends and research challenge that view.
Key Facts
- METR published in February 2026 that most developers will not work without AI on even a limited number of tasks.
- A 2025 METR study found AI-generated code slowed developers down because they spent extra time fixing errors and steering the AI.
- METR could not repeat the 2025 experiment because developers refused to work without AI for the study.
- In May 2026, METR published a survey where technical employees self-reported feeling twice as valuable with AI.
- The tokenmaxxing trend (using token count as a productivity proxy) led Amazon to shut down its internal Kirorank leaderboard after employees gamed it, running up costs.
- Uber blew through its 2026 AI budget in four months; COO Andrew Macdonald said spending had not led to a measurable increase in projects or productivity.
- Programmer and author James Shore argued AI-generated code may increase maintenance costs, writing: “You write code twice as quick now? Better hope you’ve halved your maintenance costs.”
- Entelligence AI CEO Aiswarya Sankar stated companies spend 44% of their tokens on bug fixes that their AI generated.
- Code Rabbit reported that in open source pull requests, AI produced 1.7 times more problems than human code.
- Singapore Management University researchers warned in an April report that “AI-generated code can introduce long-term maintenance costs into real software projects.”
- Cognition CEO Scott Wu (maker of AI coding agent Devin) said Devin’s skill level is between a junior and mid-level programmer.
- SMU researchers recommend programmers know what AI does well and poorly, implement quality assurance systems for AI, review AI output like a junior developer, and have humans handle architecture and security design.
Conflicting Reports
No conflicting reports identified in the source article.
Still Unclear
The article does not specify the exact financial costs of tokenmaxxing at Amazon or Uber, nor the long-term impact on overall code quality.
Misconceptions
No widespread misconceptions addressed in the source article.
Key Figures
- METR (AI research lab)
- James Shore (programmer and author)
- Aiswarya Sankar (CEO of Entelligence AI)
- Code Rabbit (code-reviewing tool company)
- Scott Wu (CEO of Cognition, maker of Devin)
- Singapore Management University (SMU) researchers
Sources: TechCrunch
