AI tools are used on a large scale by software developers, but those developers and their managers are still struggling by finding out how exactly the tools to use, with growing pins that come up on the way.
That is the collection meals of the latest survey among 49,000 professional developers through community and information hub Stackoverflow, which is strongly influenced by the addition of large language models (LLMS) to developers workflows.
The research showed that four in five developers use AI tools in their workflow in 2025 – a part that has been growing rapidly in recent years. That said: “Trust in the accuracy of AI has fallen from 40 percent in previous years to only 29 percent this year.”
The difference between those two statistics illustrates the developing and complex impact of AI tools such as Github Copilot or Cursor on the profession. There is relatively little discussion among developers that the tools are or should be useful, but people are still finding out what the best applications (and limits) are.
When asked what their top frustration was with AI tools, 45 percent of the respondents said they were struggling with “AI solutions that are almost good, but not entirely” -the largest reported problem. That is because, in contrast to output that are clearly wrong, can introduce these treacherous bugs or other problems that are difficult to identify immediately and is relatively time -consuming to solve problems, especially for junior developers who have approached work with a false sense of trust thanks to their dependence on AI.
As a result, more than a third of the developers in the survey “reports that some of their visits to stack overflow are the result of AI-related issues.” That is, codes suggestions they have accepted from an LLM -based tool, introduced problems that they then had to turn to other people to solve.
Even if major improvements have recently come through models optimized by reasoning, it is unlikely that the uncomfortable unreliability will ever disappear completely; It is endemic about the nature of how predictive technology works.
