AI Coding Tools Raise 1.7x More Problems as Developers Resist Working Without Them
Updated
Updated · TechCrunch · May 29
AI Coding Tools Raise 1.7x More Problems as Developers Resist Working Without Them
3 articles · Updated · TechCrunch · May 29
METR said it could not rerun a 2025 productivity experiment in 2026 because developers refused to do even limited tasks without AI coding tools, underscoring how dependent many have become.
That matters because earlier METR research found AI often slowed open-source developers overall: code was generated faster, but time was lost steering models, waiting for outputs, and fixing mistakes.
1.7x more problems appeared in AI-written code than in human code, according to CodeRabbit, while Entelligence AI said 44% of tokens were being spent fixing AI-generated bugs; SMU researchers separately warned of long-term maintenance costs.
Corporate spending has also raised doubts about the payoff: Amazon scrapped an internal token leaderboard after employees gamed it and drove up costs, and Uber exhausted its 2026 AI budget in four months without measurable productivity gains.
Researchers and toolmakers converge on one near-term conclusion: AI coding still needs heavy human oversight—especially for review, architecture and security—more like supervising a junior developer than replacing one.
With AI driving up costs and bugs, are developers trapped in a productivity illusion that businesses are paying for?
As AI generates more code debt, what new skills will separate elite programmers from those just managing the AI?
The AI Code Quality Crisis: Rising Defect Rates, Developer Burden, and the Future of Software Engineering (2026)
Overview
The rapid adoption of AI coding tools has created a major dilemma for software development teams. While these tools offer powerful acceleration for developers, they also introduce considerable risks to software quality. As industry studies and postmortems from late 2025 and early 2026 show, AI-authored or AI-assisted changes are often linked to more software issues, including increased bugs and slower delivery times. This has led to the cancellation of initial productivity gains that teams expected from AI integration. The result is a challenging trade-off: faster development, but with higher risks and new problems that teams must now address.