Developers Frustrated with 'Nearly-Right' AI Coding Tools, Survey Reveals
A recent global survey of software developers reveals a nuanced relationship with AI coding tools: while adoption is widespread, developer satisfaction and trust remain surprisingly low. The findings, part of Stack Overflow’s annual study, highlight significant frustrations with the current capabilities of AI in software development.
The survey, which gathered responses from 49,009 developers across 160 countries, indicates that 78.5 percent are already using AI developer tools at least monthly, with an additional 5.3 percent planning to integrate them soon. Despite this high adoption rate, sentiment is mixed. Approximately 60 percent of respondents view these tools favorably, but a notable 20 percent feel indifferent or unsure, and another 20 percent hold an unfavorable view.
A key finding is the pervasive lack of trust in AI-generated solutions. Only a mere 3.1 percent of all respondents reported “highly trusting” AI tool results. This figure drops even lower among experienced developers, to 2.5 percent, while those new to coding show slightly more faith, at 6.1 percent, though still a paltry number. Overall, 44 percent expressed some level of distrust, and even the 31 percent who were “somewhat trustful” did not exude strong confidence.
Developers identified “complex tasks” as AI’s primary weakness, with 40 percent rating its performance as “bad” or “very poor” in such scenarios. Only 4.4 percent felt AI handled complex tasks “very well,” and 17 percent avoided using AI for these tasks altogether.
Contrary to industry narratives suggesting AI is rapidly taking over code generation—such as Microsoft CEO Satya Nadella’s widely cited claim that 30 percent of Redmond’s code is AI-attributed—the survey suggests this isn’t typical across the broader software industry. Only 17 percent of developers reported mostly using AI to write code, while 29 percent stated they have no plans to use it for that purpose. Furthermore, “vibe coding,” a fully AI-centric programming method that has garnered attention, was overwhelmingly rejected, with 76 percent of respondents emphatically saying no to its use.
Instead, developers are primarily leveraging AI for more supportive roles. A significant 87 percent use AI for “searching for answers” or “learning new concepts or technologies,” indicating its value as a sophisticated search engine or educational tool.
The core frustration for many developers, cited by 66 percent, is that AI often produces “solutions that were almost right, but not quite.” This leads to further complications, as 45 percent of respondents found debugging AI-generated code to be more time-consuming than debugging human-written code. Additionally, 16 percent expressed difficulty in understanding how or why AI-generated code works.
The adoption of AI agents—a newer type of AI tool in software development—also appears limited. A substantial 69 percent of developers do not currently use agents in their workflows, with 38 percent of those having no plans to. Moreover, 41 percent reported that AI agents have had little positive impact on their productivity.
These findings underscore the enduring importance of human developers. The survey revealed that 75 percent of developers would still seek human help when they don’t trust AI answers. Human intervention is also preferred for ethical or security concerns (62 percent) and when a full understanding of a concept is desired (58 percent). Similar majorities prefer human interaction for learning best practices or when encountering difficult problems.
Despite the significant advancements and hype surrounding AI, the survey strongly suggests that human-centric workplaces will remain the norm in software development for the foreseeable future. Only a small fraction—4.3 percent of respondents—believe they will no longer need human assistance thanks to AI. This indicates that while AI tools are becoming indispensable for certain tasks, the critical thinking, problem-solving, and human judgment of developers are far from being replaced.