According to a new Stack Overflow survey of 49,009 IT professionals from 160 countries, almost all developers use AI tools to write code, but that doesn't mean they're happy with the results.
According to a new Stack Overflow survey of 49,009 IT professionals from 160 countries, almost all developers use AI tools to write code, but that doesn't mean they're happy with the results.
78.5% of respondents said they already use AI tools “monthly or infrequently.” Another 5.3% plan to start using AI this year, The Register reports .
However, within this majority, opinions are divided on the effectiveness of AI:
The Stack Overflow survey was conducted from May 29 to June 23, 2025. Respondents ranged in age from 18 to 65+, and included all types of programmers, from seasoned professionals to those just learning to code. Interestingly, however, the use of AI tools was roughly the same across all experience levels, averaging around 80%.
Only 3.1% of respondents said they had “a lot of trust” in the results of AI tools, and among experienced developers this figure is even lower at 2.5%. Beginners have the most trust in AI, with a still small 6.1% indicating high trust.
However, a cautious approach to AI seems to be the norm, with about 44% of respondents expressing “some” or “high” distrust of AI.
“Complex tasks” were AI’s biggest weakness. AI was either “bad” or “very bad” at performing complex tasks, according to 40% of respondents. Only 4.4% said the tools were “very good” at complex tasks, while 17% said they don’t use AI for complex tasks at all.
Only 17% of survey respondents said they “currently mainly” use AI for writing code, while 29% said they have no plans to use it for this purpose at all.
Regarding vibecoding, 76% answered either “no” or “absolutely not.”
Replacing or supplementing traditional search engines was one popular option, with 87% saying they use AI either to “find answers” or to “learn new concepts or technologies” (or both).
66% of developers said that AI creates “solutions that are almost right, but not quite.” Furthermore, 16% complained that it is “hard to understand how or why the code works.”
45% of respondents complain that debugging code generated by artificial intelligence “takes more time” than code written by a human.
69% of respondents said they currently do not use agents in their workflows. 38% do not plan to do so at all.
In addition, 41% of IT professionals responded that agents had a very slight positive impact on their productivity.
75% of developers would still seek help from a human in cases "when I don't trust the AI's answers."
62% believe that ethical or security issues with code require human intervention.
58% would turn to a person "when I want to fully understand something."
And only 4.3% of respondents said, “I don’t think I will need help from people anymore,” thanks to artificial intelligence.
