Coders are refusing to work without AI — and that could come back to bite them

In a startling trend observed in 2026, developers have become increasingly reliant on artificial intelligence (AI) coding tools, with researchers at METR revealing that many are unwilling to work on tasks without the assistance of AI. While AI is known to accelerate coding speed, concerns are emerging regarding the quality of the code produced, which could lead to significant challenges in the future. METR initially aimed to build upon previous research from 2025 that assessed the productivity of open-source developers utilizing AI. Although developers expressed feelings of increased productivity, they were surprised to find that the actual time spent on tasks often increased due to the need for error correction and guidance of AI systems. When METR attempted to replicate this study, they encountered a barrier: developers refused to participate without AI support, highlighting their dependency on these tools. To gather insights, METR later conducted a survey where technical staff self-reported their productivity enhancements attributed to AI. The results indicated that many believe AI has significantly increased their value within their organizations. However, recent reports have cast doubt on these self-assessments, particularly in light of escalating costs associated with AI usage. The practice known as 'tokenmaxxing'—tracking productivity through token usage—has gained traction in 2026, but it appears to be waning. Amazon's recent closure of its internal token leaderboard, Kirorank, due to employees gaming the system with excessive AI usage, underscores the disconnect between AI use and actual productivity gains. Similarly, Uber's rapid depletion of its AI budget within the year's first four months, without a corresponding increase in output, raises eyebrows regarding the effectiveness of AI tools. Furthermore, AI-generated code may not alleviate ongoing maintenance burdens; instead, it could exacerbate them. James Shore, a programmer and author, warned that developers must be cautious: a temporary boost in coding speed does not guarantee lower maintenance costs. Evidence suggests that companies are allocating a significant amount of their AI resources to fix bugs generated by AI, highlighting a critical flaw in the technology. Recent analysis by Code Rabbit indicates that AI-produced code leads to 1.7 times more issues than code written by humans. This concern is echoed by independent research from Singapore Management University, which cautioned that AI-generated code could impose long-term maintenance costs on software projects. As developers grow fond of their AI assistants, the question remains: what is the way forward? Some advocates for AI tools propose that developers could delegate the laborious task of fixing code to AI agents. However, even proponents like Cognition's CEO Scott Wu admit that current AI capabilities fall short of those of an experienced developer. The Singapore Management University researchers recommend a more nuanced approach: developers should become intimately familiar with the strengths and limitations of AI tools, ensuring robust quality assurance processes are in place. They emphasize the importance of human oversight in areas like software architecture and security design, suggesting that while AI can assist, it cannot replace the critical thinking and strategic planning that experienced developers provide.