Developers are increasingly refusing to work without AI

In February 2026, METR, an AI research lab, published a revelation about developer habits: most developers refuse to work without AI, even on limited tasks. The lab had planned to update groundbreaking research published in 2025 regarding AI coding productivity. That earlier study measured how much time open-source developers took to complete tasks by hand versus with AI. However, METR could not repeat the experiment because developers were unwilling to participate without their AI assistants. Instead, in May, METR published a survey on AI productivity gains where technical employees self-reported their performance. While these developers perceived that AI made them more valuable, other data suggests this self-reported productivity may not align with actual software quality.

This shift coincides with growing scrutiny over “tokenmaxxing”—the trend of using the number of tokens used as a proxy for AI productivity. This trend may be over as companies face rising costs and limited returns. Amazon shut down Kirorank, its internal token-tracking leaderboard, after employees reportedly gamed the system by using AI agents excessively. Meanwhile, Uber reported budget issues related to AI spending, blowing through its budget early in the year. Data from the industry supports these concerns. Aiswarya Sankar, founder and CEO of Entelligence AI, asserted that companies are spending 44% of their tokens on bug fixes generated by AI. Additionally, code-reviewing tool company CodeRabbit reported that AI produced 1.7x more problems than human code.

These findings highlight a deeper issue: AI-generated code may not be better code and could cause future problems. James Shore, a programmer and author, argued that AI-generated code does not necessarily reduce maintenance needs and may actually increase them. Shore warned: “You write code twice as quick now? Better hope you’ve halved your maintenance costs. Otherwise, you’re screwed. You’re trading a temporary speed boost for permanent indenture.” Research from Singapore Management University supports this warning. In April, the university published a report asserting that AI-generated code can introduce long-term maintenance costs into real software projects. Even as Cognition founder and CEO Scott Wu builds AI coding agents like Devin, researchers warn that organizations must carefully manage the long-term risks of automated code.

Why it matters

Organizations face a potential productivity trap where the speed gains of AI coding are offset by increased long-term maintenance burdens and technical debt.