Coders refuse to work without AI – and that can come back to bite them

By 2026, you won’t be able to release AI coding tools without holding an engineer’s vise, researchers have found.

But while AI is undoubtedly helping coders produce code faster, it may not produce better code, some researchers warn. And that can cause them problems down the road.

Specifically, in February 2026, the respected AI research lab METR published a surprising revelation: the majority of engineers will not work, even in a limited number of tasks, without AI.

METR was hoping to provide an update on some alarming research published a few months earlier, in 2025, about AI code generation. In it, researchers measured how much time it took open source developers to perform tasks by hand versus AI.

While developers in that study reported that AI made them more productive, they were shocked to learn that it actually slowed them down. Sure, it generated code faster, but then they spent more time finding and fixing bugs, guiding the AI ​​and waiting on it to complete tasks.

When METR set out to repeat the experiment to measure improvements in AI and coding skills, it couldn’t.

Devs were reluctant to participate “because they don’t want to work without AI” even just for research, the researchers admitted.

Instead, METR published a survey in May that allowed tech workers to self-report their benefits of AI productivity. No wonder they see that AI has made them twice as valuable to their organizations.

But recent headlines about the wild costs of so-called tokenmaxxing, coupled with conflicting recent research, cast such notions into question.

Tokenmaxxing, or using the number of tokens a person spends as a proxy for AI production, has been the trend of 2026 so far. And it may be over.

Amazon shut down an internal token-tracking leaderboard called Kirorank after employees were gaming it by using AI agents excessively, and raising costs, the Financial Times reported this week. Employees have proven that the use of AI does not automatically translate into increased productivity.

Uber blew its AI budget for 2026 during the first four months of the year, the report said. COO Andrew Macdonald recently said on a podcast that such spending has not led to a measurable increase in projects or productivity.

AI-generated code also does not reduce ongoing code maintenance needs, and may even increase them, programmer and author James Shore argued well in a blog post published by Hacker News.

“Coding twice as fast now? We better hope you cut your maintenance costs in half,” he wrote. “Otherwise, you’re corrupt. You’re trading a temporary speed boost for permanent independence.”

There is some evidence that AI can increase the burden of code maintenance.

A viral tweet from Aiswarya Sankar, founder and CEO of trusted agent engineering startup Entelligence AI, announces that companies are spending 44% of their tokens on fixing bugs generated by their AI. Meanwhile, code analysis tools company Code Rabbit says it analyzed open source pull requests and found that AI produced 1.7x more problems than human code.

Those are, admittedly, self-serving statistics from those trying to sell AI code review tools.

However, independent researchers have also found such stories. Researchers from the prestigious Singapore Management University published a report in April warning that “AI-generated code can introduce long-term maintenance costs to real-world software projects.”

Given that programmers love their AI assistants, what’s the solution?

Well, those who want to sell you AI coding agents say that devs can just use AI coding agents to do the bone-crunching tasks of fixing code as fast as the AI ​​spits it out. That’s what Cognition founder and CEO Scott Wu — maker of AI code agent Devin — suggests.

But he also admits that, although Devin can work independently, he can currently rate his skills between a high- and mid-level designer, depending on the task. This is not a shut up and forget about the solution.

SMU researchers suggest a more human approach. Programmers should know what tasks AI does and doesn’t do well as deeply as they know their favorite coding languages. They need robust quality assurance systems designed for AI and are still stuck reviewing AI work as if it were a junior dev.

Meanwhile, the researchers say (and Wu agrees), people still have to do big-picture work like software development and security design.