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

A growing share of professional software developers now consider AI coding assistants non-negotiable, according to researchers tracking adoption trends in the tech industry, even as separate studies raise doubts about whether the resulting code is actually any better.

The starkest evidence comes from METR, a respected AI research organization, which published findings in February 2026 showing that a majority of programmers surveyed were unwilling to complete even a small set of coding assignments without leaning on an AI tool. The finding suggests that AI assistance has shifted from a helpful convenience to something many developers treat as essential infrastructure for getting their jobs done.

That level of dependency marks a notable shift from just a few years ago, when AI coding tools were still viewed by many engineers as experimental add-ons best suited for boilerplate work or quick prototyping. Now, according to the researchers, refusing an assignment altogether has become a more common reaction than simply working without the technology.

Yet speed is not the same as quality, and that distinction is at the center of a separate set of warnings from other researchers. While AI tools are widely credited with helping programmers churn out code more quickly, evidence that the resulting output is more reliable, more maintainable, or less error-prone remains thin. Some researchers caution that this gap between velocity and quality could create hidden costs that surface only later, once flawed or poorly understood code makes its way deeper into production systems.

The concern extends beyond individual bugs. If engineers are leaning on AI to generate large volumes of code without fully internalizing how that code works, teams may end up with software that is harder to debug, audit, or extend down the line. That risk is compounded by the fact that many developers now say they are reluctant to even attempt tasks without AI support, raising questions about how core coding skills might erode if the tools become unavailable or unreliable.

Not everyone frames the trend as alarming. Proponents of AI-assisted development argue that offloading routine or repetitive coding work frees engineers to focus on architecture, testing, and other higher-level decisions that still require human judgment. From that vantage point, resistance to coding without AI support could simply reflect a rational response to a tool that meaningfully boosts output, rather than a sign of eroding competence.

Still, the tension between those two narratives — faster output versus uncertain code quality — is likely to shape how companies approach AI tooling policies in the months ahead. Engineering leaders may face pressure to introduce more rigorous review processes, quality benchmarks, or training safeguards to ensure that speed gains from AI assistance do not come at the expense of long-term software reliability.

Researchers say further study is needed to determine whether the productivity gains attributed to AI coding tools hold up once the software they help create is tested over time in real-world conditions, rather than judged solely on how quickly it was written.

Source: TechCrunch