The 2025 Stack Overflow Developer Survey offers a window into how developers really feel about AI. Adoption of AI tools is booming, yet many coders remain cautious about trust, complexity and the parts of the workflow where AI is invited. This post pulls together the key points from the survey’s AI → Developer Tools section.
AI tools are everywhere—but trust is low
The survey shows that almost everyone is using or at least planning to use AI in their work. About four out of five respondents said they use or plan to use AI tools, and more than half of professional developers use AI every day. Early‑career developers (those with 1–5 years of experience) are the most enthusiastic—over half of them reach for AI tools daily.
Despite the rapid adoption, positive feelings toward AI have cooled. This year only about 60 percent of respondents described their sentiment toward AI tools as favorable or very favorable. That’s down from over 70 percent in 2023 and 2024. Developers also don’t blindly trust AI’s output. Only about three percent of respondents said they “highly trust” the accuracy of AI tools, while nearly half expressed some level of distrust. Seasoned developers are the most skeptical; they reported the lowest rate of high trust and the highest rate of high distrust.
AI handles research and documentation better than complex tasks
Last year 35 percent of professional developers said AI struggled with complex tasks. This year that number has dropped slightly to 29 percent, but very few people believe AI truly excels at complexity. Only around four percent think AI tools handle complex coding tasks very well. Many respondents said AI is “good but not great,” and a significant portion said AI is outright bad or very poor at handling complex work. In other words, AI still isn’t ready to own mission‑critical or high‑risk tasks without human oversight.
Where developers are using AI today
When asked which parts of the workflow they use AI for, developers pointed to research, learning and documentation as the biggest areas. Here’s where AI shows up most often:
- Searching for answers – Over half of heavy AI users rely on it to look up information and troubleshoot issues. Partial users lean on it here as well.
- Generating content or synthetic data – AI is popular for creating content like sample datasets or boilerplate text. About one‑third of heavy users and one‑third of partial users do this.
- Learning new concepts and technologies – Many respondents turn to AI to understand unfamiliar languages, frameworks or techniques.
- Documenting code and creating documentation – Around one‑third use AI to help write comments and documentation for projects.
- Understanding a codebase and debugging – Some developers use AI to explore existing code or assist with debugging, although this is more common among partial users than heavy users.
- Writing and testing code – Only a small share of heavy users say they mostly write code with AI, but a much larger share use it to write or test code partially.
What the future looks like
Developers were also asked about their plans over the next three to five years. Many said they plan to partially use AI for learning about a codebase, testing and writing code. Roughly a quarter are considering AI for predictive analytics, project planning and even deployment or monitoring. A smaller group plans to mostly use AI for generating content and documentation.
However, there are still areas where developers want to keep AI at arm’s length. The survey notes that people show the most resistance to using AI for high‑responsibility tasks like deployment, monitoring and project planning. More than two‑thirds of respondents do not plan to use AI for project planning, and three‑quarters don’t want AI handling deployment or monitoring.
Satisfaction depends on how you use AI
Interestingly, satisfaction with AI tools correlates with how much you use them. Those who currently use AI for most of their tasks tend to be quite satisfied. They report using AI frequently for searching and learning. People who say they plan to rely heavily on AI in the future are a little less satisfied with the tools available today, which suggests that expectations are high and current solutions haven’t fully met them yet.
Top frustrations
AI isn’t perfect, and the survey highlights several pain points. The biggest frustration, cited by about two‑thirds of developers, is getting answers or code that are “almost right, but not quite.” This leads straight into the second‑biggest issue: debugging AI‑generated code takes longer than doing it yourself. Other annoyances include not using AI regularly enough to build trust, feeling less confident in one’s own problem‑solving and having trouble understanding how AI‑generated code works. Only a tiny minority said they haven’t encountered any problems.
People still matter
Even if AI becomes capable of writing and testing most of our code, developers are clear about one thing: they still want other humans in the loop. When asked about a future where AI handles most coding tasks, three‑quarters of respondents said they would still seek human help when they don’t trust AI’s answers. Many also cited ethical or security concerns, a desire to fully understand or learn best practices, being stuck on a problem, or needing to fix complex or unfamiliar code. Only about four percent imagine they won’t need help from people at all.
“Vibe coding” is niche
The survey also asked about “vibe coding,” which refers to generating software entirely from natural‑language prompts. The answer? Most developers aren’t doing it. About 72 percent said vibe coding is not part of their professional workflow, and only around 12 percent said they do it. A few respondents have tried it or use it somewhat, but the overwhelming majority either avoid it or dismiss it.
Final thoughts
The 2025 Stack Overflow survey paints a nuanced picture of AI in software development. On the one hand, AI tools are becoming as common as IDEs or version control: they help developers find answers, learn new things and take care of repetitive documentation tasks. On the other hand, trust and satisfaction lag behind adoption. Developers remain skeptical about the accuracy of AI output, worry about its ability to handle complex tasks and are reluctant to let AI take over high‑risk responsibilities like deployment and project planning.
Perhaps the most important takeaway is that people still matter. No matter how sophisticated our tools get, developers plan to lean on each other for advice, sanity‑checks and ethical guidance. AI can make us faster and more efficient, but it hasn’t replaced the need for human judgment and collaboration. That’s a reassuring conclusion for anyone worried that robots are about to code us out of a job.
If you enjoyed this write‑up, you can find more of my work on my personal site: takiuddin.me.
Top comments (0)