- Developers are less likely to rely on AI-Generated Output than they were last year, report requirements
- Only 2.5% of the experienced developers trust AI -Output a lot
- AI is great to help new developers learn to code
As developers become more accustomed to AI tools in their workflows, it becomes increasingly clear that they do not always have confidence in its output, new research has claimed.
The latest developer survey from Stack Overflow has revealed, although the AI adoption is up to 84% from 76% by 2024, Ther has also been a major leap in the number of developers who do not trust AI-generated results, up from 31% in 2024 to 46% in 2025.
On the flip side, only 3.1%trust AI results – a mood that is more common among beginners (6.1%) than it is among experienced DEVS (2.5%).
Developers are not sure of AI, still
Currently, as many as 78.5% of developers use AI on rare basis, such as monthly, and this is consistent across all levels of experience according to the study.
Despite the lack of trust, it is clear that developers see artificial intelligence as a useful starting point, with as many as three out of five who see AI tools, positively compared to only one in five who see it unfavorable (and one more one in five who don’t care).
But that’s all it is at the moment – a starting point. Three-quarters admitted that they would still ask a human when they do not trust AI response, with 58% prefer to ask people when they do not fully understand something and a similar number seeking human help with ethical and security concerns.
“AI is a powerful tool, but it has significant risks of incorrect information or may lack complexity or relevance,” explained Stack Overflow CEO Prashanth Chandraskar.
While its use of cases in the development cycle may be more limited, artificial intelligence shows to be useful in other areas – 44% use it to learn to code (up from 37% last year) and 36% use it for work or progress.
“By providing a trusted human intelligence layer in AI’s age, we believe that the technical enthusiasts today can play a greater role in adding value to building AI technologies and products from tomorrow,” CHANDRASSKAR added.
