In 2024, software engineer Laura Housh started using AI as a helper for coding and other tasks. Two years on, she describes the relationship as love-hate, leaning complicated. “I’ve tried to use it for coding, and I just can’t get behind it,” she told CNN. The tools can point you in a direction or serve as a second pair of eyes, she said, but they lack the context to be truly useful. She has company.
Six in 10 tech workers in a new Stack Overflow survey say they try to avoid using AI. The survey — 103 questions, 30,000 software developers in 169 countries, shared exclusively with CNN before its Tuesday publication — is one of the better windows into what happens inside the industry that got AI first. An Indeed Hiring Lab report last year ranked “software developer” as the job most poised to be transformed by AI, and the transformation is well underway: both tech giants and startups increasingly use AI to write code.
Now, you may notice a small tension in the numbers. Sixty percent try to avoid AI, yet the same survey found that 65% of respondents use coding assistants or agents. And last year Google’s research division surveyed 5,000 tech professionals worldwide and found that 90% use AI in their jobs, a 14% increase from 2024. This is less paradoxical than it looks. “Use” and “embrace” are different verbs. If your employer rolls out a coding assistant, your performance review starts assuming one, and the industry announces weekly that the tools are mandatory, you can use the thing every day while quietly not trusting it — the way you might use a junior colleague who is fast, confident and occasionally makes things up.
The mood data are unambiguous about where this lands. Nearly one in three respondents say they are not happy in their current roles, up from 28.4% last year. About 45% describe themselves as “complacent” and only 22.3% as “happy.” The biggest drivers among the complacent and unhappy are burnout, tech fatigue, economic uncertainty and a lack of clarity about how to get promoted — complaints that would exist in any economy. But the pace of AI has, in the word of Stack Overflow CEO Prashanth Chandrasekar, “augmented” them. “People have to completely retool their jobs, and so it is quite unnerving for folks to know exactly how much of what they’re doing will be automated,” he told CNN.
Chandrasekar sketches a squeeze that hits at both ends of the career ladder. Senior programmers may find that the skills they spent years honing no longer command what they did. Entry-level workers are now expected to perform as if they already have three years of experience, because, as he put it, “the bar has been raised” by AI. And everyone is meant to keep up in their spare time. Housh says she watches AI videos and reads platform documentation outside work and still struggles. “It’s taxing because you have a lot of noise saying, ‘Do this, do this, do this, or you’ll get left behind,’” she said.
For engineers like Housh and Audrey Eschright, who has 20 years in the field, the objection is more precise than fear of replacement. An AI agent has neither the broad general knowledge of an experienced developer nor the job-specific knowledge of a longtime employee, they argue — and delegating the coding can leave the human out of the loop of the solution. “Writing code is how you solve problems,” Eschright said. “As a software engineer, the process of figuring out what a system does is to change it.” It is the same argument we heard from a veteran engineer about why code review survives the coding agents: the typing was never the job.
The survey lets you watch developers draw that line in real time. A little more than 74% say AI is useful for implementing code and about 72% for debugging — the grinding parts. But only 34.4% find it useful for other job functions like communication and design. Just 6.6% trust AI with important decisions, and about 78% say source attribution matters when checking AI-generated code. This caution is arriving just as the stakes rise: a January Deloitte report found that 54% of the 3,200-plus business and IT leaders surveyed expected to move 40% or more of their AI experiments from pilot to production within three to six months, which is exactly why Chandrasekar says the trust question is sharper now — the code is going live.
Housh, for her part, has settled on a posture rather than a boycott. “I don’t like losing control; I like critical thinking and curiosity,” she said. “It’s just so important to be more curious than ever.” Sixty percent of her peers are trying to avoid the tool and 65% are using it, and those two facts will coexist in many of the same people for years. The robots are not taking the job; they are joining it, awkwardly, like everyone else who was hired in a hurry.