In 2026, developers are clinging to AI coding tools so tightly that researchers can't pry them away. While these tools undoubtedly help programmers write code faster, some experts warn that faster doesn't necessarily mean better—and this could create serious problems down the line.

Back in February 2026, the respected AI research lab METR dropped a surprising finding: most developers now refuse to work without AI, even on a small number of tasks. METR had planned to update groundbreaking research from 2025, which measured how long open source developers took to complete tasks manually versus with AI assistance. In that earlier study, developers reported feeling more productive with AI, but were shocked to learn it actually slowed them down. The AI generated code quickly, but developers spent extra time hunting for errors, steering the model, and waiting for it to finish tasks.

When METR tried to repeat the experiment to see how AI and coder proficiency had improved, they hit a wall. Developers wouldn't participate "because they do not wish to work without AI"—even for the study itself, the researchers admitted. Instead, METR published a survey in May where technical employees self-reported productivity gains. Unsurprisingly, they felt AI made them twice as valuable to their organizations. But recent headlines about the skyrocketing cost of "tokenmaxxing," along with a handful of new studies, cast doubt on those self-perceptions.

Tokenmaxxing—using the number of tokens consumed as a proxy for AI productivity—has been the trend of 2026 so far. And it may already be fizzling out. Amazon shut down its internal token-tracking leaderboard, Kirorank, after employees gamed the system by using AI agents excessively, driving up costs, according to the Financial Times. The employees proved that heavy AI use doesn't automatically mean higher productivity. Uber burned through its entire 2026 AI budget in just four months, The Information reported. COO Andrew Macdonald recently admitted on a podcast that the spending hadn't led to a measurable increase in projects or productivity.

AI-generated code also doesn't reduce ongoing maintenance needs—and may increase them. Programmer and author James Shore made this point elegantly in a viral Hacker News blog post. "You write code twice as quick now. Better hope you've halved your maintenance costs," he wrote. "Otherwise, you're screwed. You're trading a temporary speed boost for permanent indenture."

Other evidence backs him up. A viral tweet from Aiswarya Sankar, founder and CEO of reliability engineering startup Entelligence AI, claims companies are spending 44% of their tokens on fixing bugs that AI itself created. Code review tool company Code Rabbit says it analyzed open source pull requests and found AI produced 1.7 times more problems than human-written code. These numbers admittedly come from companies selling AI code review tools, but independent researchers have found similar issues. A report from Singapore Management University, published in April, warned that "AI-generated code can introduce long-term maintenance costs into real software projects."

So if programmers love their AI assistants, what's the fix? Companies selling AI coding agents suggest developers can simply use those agents to handle the tedious work of fixing code as fast as AI generates it. Scott Wu, founder and CEO of Cognition and creator of the AI coding agent Devin, advocates this approach. But even he admits that while Devin can work independently, he'd currently rate its skill between a junior and mid-level programmer, depending on the task. It's not a set-it-and-forget-it solution.

The SMU researchers propose a more human-centered approach. Programmers should know what tasks AI handles well and poorly as deeply as they know their favorite coding languages. They need strong quality assurance systems designed specifically for AI, and they must carefully review AI output as if it came from a junior developer. Meanwhile, the researchers say—and Wu agrees—humans should still handle the big-picture work, like software architecture and security design.