Coders refuse to work without AI, raising long-term concerns

The Story

According to a TechCrunch report citing research from AI lab METR, a majority of developers will not work without AI tools, even for limited tasks. METR attempted to replicate a 2025 study on AI coding productivity but found that developers refused to participate because they did not wish to work without AI. Instead, METR published a survey in May 2026 in which technical employees self-reported that AI made them twice as valuable. However, the article notes recent headlines about the expense of “tokenmaxxing” and research that questions whether AI actually improves productivity or reduces maintenance costs. Amazon shut down its internal token-tracking leaderboard after employees were gaming it, and Uber blew through its 2026 AI budget in four months without measurable gains. Researchers from Singapore Management University warned in April that AI-generated code can introduce long-term maintenance costs. Despite these concerns, AI coding tools remain popular among developers.

Key Facts

• METR reported in February 2026 that most developers refuse to work without AI, even on limited tasks.
• METR could not repeat a 2025 study because developers would not participate without AI.
• A May 2026 survey from METR found technical employees perceived AI made them twice as valuable.
• Amazon shut down its token-tracking leaderboard, Kirorank, after employees were using excessive AI agents and running up costs.
• Uber exhausted its 2026 AI budget in four months, and COO Andrew Macdonald said spending had not led to a measurable increase in projects or productivity.
• Programmer James Shore argued that faster AI code generation does not necessarily reduce maintenance costs.
• A tweet from Entelligence AI CEO Aiswarya Sankar claimed companies spend 44% of tokens on bug fixes from AI-generated code.
• CodeRabbit found AI produced 1.7 times more problems than human code in open source pull requests.
• Singapore Management University researchers warned in April that AI-generated code can introduce long-term maintenance costs.
• Cognition CEO Scott Wu said his AI coding agent Devin currently performs between a junior and mid-level programmer.

Conflicting Reports

No conflicting reports identified in the source article.

Still Unclear

Whether the perceived productivity gains from AI translate into actual, measurable improvements in code quality or reduced long-term costs.

Misconceptions

The article addresses the misconception that AI coding tools automatically increase productivity, noting that they can increase time spent on fixing errors, waiting on AI, and managing maintenance.

Key Figures

• METR (AI research lab)
• Andrew Macdonald (COO, Uber)
• James Shore (programmer and author)
• Aiswarya Sankar (CEO, Entelligence AI)
• CodeRabbit (code-reviewing tool company)
• Singapore Management University (researchers)
• Scott Wu (CEO, Cognition)

Sources: TechCrunch