Stack Overflow survey shows developers trust AI with caveats

Stack Overflow has published the results of its 2026 Developer Survey, which found that most developers use AI tools at work.

The survey drew more than 30,000 responses from technologists in 169 countries and examined attitudes to AI, workplace conditions, learning habits, and technology choices.

The findings suggest developer sentiment towards AI has become more measured, rather than fully sceptical or fully trusting. Nearly half of respondents said they trust AI output when they can verify it easily, while only 6.6% said they would trust it for important decisions.

Source attribution also emerged as a central concern. Some 78.7% of respondents said attribution was important or very important when assessing AI-generated code or technical answers.

"Developers are historically a highly skeptical group, and over the last year, as AI agents become rapidly more autonomous, that skepticism now acts as a critical safeguard," said Prashanth Chandrasekar, Chief Executive Officer of Stack Overflow.

He added: "As agents take on more work, developers need to know where an answer came from, what context informed the information that was produced or a decision that was made, and, most crucially, that they can verify its accuracy and relevance. Trust is not a given when it comes to developers; it is earned through transparency, source attribution, and rich technical context."

AI at work

AI tool use at work appears to be widespread. Coding assistants or coding agents were used by 66% of respondents, while 63% said they used general-purpose chat tools and 26.2% used automated agent workflows. Only 17.2% said they did not use AI tools.

The data also points to heavier use among a core group of regular users. Among daily AI users, 31% said they spent four hours or more a day using the tools, and 80% said they used AI for at least one hour each day.

Developers were most likely to say AI was useful for tasks closest to software creation itself. Some 74% found it useful or very useful for implementation, while 66% said the same for debugging or refactoring code. That figure dropped for team communication and design, where around a third to a half of respondents reported it useful.

Governance appears to lag adoption. Three in ten respondents said AI use at work was optional and left to the individual, while only 24% said their employer had published approved tools or formal AI guidance.

Among developers who changed AI tools, 25% said output quality was the main reason, while 11% said they switched because their organisation required it. At the same time, a significant minority remain resistant, with concerns about skill atrophy, replacement, and ethics among the leading reasons.

Workplace pressure

Beyond AI, the survey paints a picture of a profession dealing with uncertainty at work. It found that 45% of developers were classified as complacent in their current jobs, compared with 33% who were unhappy and 22% who were happy.

Among those described as complacent, burnout or tech fatigue was the most commonly cited cause at 21%. Unclear priorities were the biggest source of workplace friction at 17%, followed by inefficient workflows at 15% and insufficient time or resources at 14%.

Employment patterns are also shifting. Freelancers and sole proprietors made up 11% of respondents, up from 4% a year earlier.

Remote work remained the most common arrangement, accounting for 35% of respondents. That compared with 20% in hybrid roles leaning towards in-person attendance and 18% working fully in person.

Despite these pressures, learning activity remained strong. More than half of respondents, 52%, said they had acquired a new coding skill or learned a new programming language over the year.

Technical documentation was the most common learning source for that group at 58.9%. AI code-generation tools were used by 52.7%, other online resources by 51.7%, and Stack Overflow itself by 33.8%.

Tools and choices

The survey also showed that developers continue to rely on familiar technologies even as AI reshapes workflows. JavaScript remained the most widely used language at 62%, followed by SQL at 58.4%, HTML and CSS at 58.3%, and Python at 58.1%.

Python and JavaScript were also the most common languages used alongside AI tools, at 39% and 38% respectively. In databases, PostgreSQL led with 58% usage over the past year, ahead of SQLite at 42% and MySQL at 35%.

When developers recommend technologies, AI features rank low on the list. Reliability, performance, and stability were the top reasons at 17%, ahead of productivity or time savings at 16% and community or ecosystem strength at 13%. AI features were cited by 2%.

For problem-solving, online search remained the most common first step at 83%, followed by an AI agent at 70%. Maintained repositories were selected by 44%, while 29% said they would directly contact a trusted person.

The findings also suggest close competition among large language model providers. Anthropic and OpenAI recorded current usage levels of 75% and 74% respectively, while 49.0% of respondents said they wanted to work with Anthropic models in the coming year, compared with 32% for OpenAI.

Among community resources, public GitHub projects narrowly led at 69.5%, with Stack Overflow close behind at 68.6%, followed by YouTube at 58% and Reddit at 54%.