For Programmers, Even As Ai Adoption Climbs, Trust Wanes

fotograzia/Getty Images

Programmers are utilizing AI more than ever, but they don't for illustration aliases spot nan devices very much, according to the 2025 Stack Overflow Developer Survey.

The study of almost 50,000 developers recovered that 84% now usage aliases scheme to usage AI devices successful their workflow, up from past twelvemonth (76%). Over half of master developers (51%) usage these devices daily. 

Also: The champion AI for coding successful 2025 (and what not to use)

Such figures mightiness propose that programmers must emotion AI. However, only 60% expressed affirmative sentiment toward AI tools, a proportionality down from complete 70% successful some 2023 and 2024.

Distrust is simply a defining taxable of nan survey. In 2024, 43% of developers felt bully astir AI accuracy, and only 31% were skeptical. By 2025, 33% of developers trusted AI instrumentality outputs, 46% expressed progressive distrust, and a specified 3% said they highly trusted nan results. Among seasoned professionals, nan "highly trust" fig dropped to conscionable 2.6%, pinch 20% reporting beardown skepticism. 

Also: Bad vibes: How an AI supplier coded its measurement to disaster

In short, developers surely usage AI, but trusting nan exertion to get nan occupation correct connected its ain is different matter. And aft IT leader Jason Lemkin's acquisition pinch a Vibe programming task that disintegrated, taking pinch it a accumulation database, who tin blasted them? 

The Stack Overflow study besides recovered that nan biggest azygous frustration, cited by 66% of developers, is dealing pinch "AI solutions that are almost right, but not quite," which often leads to nan second-biggest frustration, astatine 45%, being "Debugging AI-generated codification is much time-consuming."

As Bill Harding, CEO of Amplenote and GitClear, noted successful GitClear's AI Copilot Code Quality study, which analyzed 211 cardinal lines of code, "developers spot nan existent procreation of AI assistants astir arsenic overmuch arsenic we trusted nan erstwhile generation, i.e., not much." 

It's not conscionable programmers who don't spot AI. A caller study of much than 1,100 Americans revealed that only 8.5% said they "always trust" nan accusation they person from Google's AI Overviews, and 21% said they person zero spot successful nan feature's expertise to aboveground meticulous information. A caller KPMG study recovered that, worldwide, 66% of group usage AI, but only 46% spot AI systems.

Also: How to get free of AI Overviews successful Google Search: 4 easy ways

Junior developers who spot AI nan astir could beryllium making a large mistake. A celebrated blog station by Namanyay Goel, an independent developer, warned: "We're trading heavy knowing for speedy fixes, and while it feels awesome successful nan moment, we're going to salary for this later." 

Worries complete AI-created method indebtedness are growing. Harding warned that if companies proceed to measurement developer productivity by nan simple-minded metric of number of commits aliases lines of codification written, AI-driven method indebtedness will spiral retired of control. "Leaders request to admit that much codification is often worse," he said, suggesting that copying and pasting codification leads to higher defect rates. Indeed, GitClear recovered a nonstop relationship betwixt nan rising defect complaint and AI adoption.

All that said, Stack Overflow besides revealed successful its study that OpenAI's GPT models are nan astir celebrated ample connection models, pinch 82% of developers who usage AI indicating that they utilized them for improvement activity successful nan past year. Anthropic's Claude Sonnet models came second, followed by Google's Gemini Flash.

Despite AI's rise, erstwhile it comes to integrated improvement environments (IDEs), programmers still for illustration Visual Studio (75%) and Visual Studio Code (29%) complete AI-first programming IDEs. The old-school, elemental codification editors, Vim and Notepad++, stay popular, moreover among programmers who usage AI. That said, Microsoft's incorporation of Copilot into its tools has proven to beryllium a smart move.

Also: Coding pinch AI? My apical 5 tips for vetting its output - and staying retired of trouble

As before, JavaScript, HTML/CSS, and Python support their position arsenic nan astir wide utilized programming languages. Python, and astir apt owed to nan fame of Python-based generative AI libraries TensorFlow and PyTorch, is particularly sought aft by developers adopting a caller language. However, Rust, pinch an 83% support mark, remains nan astir admired language.

Looking ahead, while AI is being adopted quickly, developers are, if anything, increasing cautious astir handing disconnected captious tasks to agents. A immense mostly (75%) said quality proposal is still irreplaceable successful scenarios wherever they don't spot AI's output. 

As for AI agents, they person yet to scope mainstream acceptance. Over half of nan survey's respondents usage simpler AI tools, and 38% person nary plans to adopt agents soon.