A review by Stack Overflow of data from the 2024 and 2025 developer surveys shows that AI has become a widespread part of workflows, but increased use has not been accompanied by a comparable rise in enthusiasm or trust. The proportion of developers using AI tools rose from 62% in 2024 to 79% in 2025, and about half of respondents said they use them daily.
The review draws on 15 years of Developer Survey data, prior to the release of the 2026 survey results, and focuses on three interconnected areas: the evolution of AI, working conditions, and then demographics and the role of the technical community.
From Coding Assistants to Agents
Stack Overflow began measuring the use of AI agents in its 2025 survey, with 31% of respondents reporting that they used them. ChatGPT and GitHub Copilot were the most prominent gateways to AI tools, at 82% and 68%, respectively, among users of ready-made assistants, followed by Google Gemini at 47%, then Claude Code at 41% and Microsoft Copilot at 31%.
In a smaller pulse survey conducted in April 2026, agent usage rose to approximately 59%, while Claude Code’s share also increased from 41% to 55%. The review indicates that usage was more strongly associated with daily users and people in executive roles, without this meaning that agents have become fully autonomous in software engineering work.
The most common uses remain related to discovery, drafting, and assistance. In the 2024 data, 82% of respondents used AI to write code, 68% for searching, 57% for debugging, and 40% for preparing documentation. The analysis cautions that direct comparison with 2025 is limited because of changes in question design, but the overall trend indicates that programming is still mostly AI-assisted rather than fully delegated to AI.
Learners Use the Tools More and Trust Them Less
The proportion of people learning to program who use AI tools rose from 63.0% in 2024 to 73.3% in 2025, including 39.5% who use them daily. By contrast, positive views of these tools fell from 72.2% to 52.8%, while skeptical views rose from 6.4% to 26.3%.
The proportion of students expressing medium or high confidence in the accuracy of outputs also declined from 49.1% to 37.4%, and positive assessments of using AI for complex tasks fell from 41.5% to 33.7%. The source indicates that these figures do not prove that AI harms learning, but they highlight an important distinction between producing code that works and understanding why it works.
Agents did not receive broad acceptance among students; 28.4% used them in 2025, while 44.1% said they did not intend to adopt them. Concern about job replacement also rose among all developers, from 12.1% in 2024 to 15.0% in 2025.
Work Environments Become More Complex
The proportion of fully remote work declined from 38.0% in 2024 to 32.4% in 2025, while fully in-person work reached 17.9% and structured hybrid work models reached 37.1%. A total of 12.6% said they had substantial flexibility, allowing them to attend according to their preferences or project requirements. Using a different calculation, the proportion of people enjoying some degree of remote-work flexibility reached 82% in 2025, compared with 80% in 2024.
Geographical differences show that the United States recorded the highest rate of fully remote work, at 45.0%, while India recorded the highest rate of in-person work, at 30.5%. In Germany, the category of flexible work in which employees choose when to attend reached 20.7%, alongside 24.6% for hybrid work leaning toward flexibility.
Job satisfaction remained distributed among 28.4% who were dissatisfied, 47.1% in the middle, and 24.5% who were satisfied. Autonomy and confidence in managing tasks topped the satisfaction factors in 2025, ahead of pay and benefits and solving real-world problems. At the same time, 85.3% of respondents were individual contributors, compared with 14.7% in people management.
What Does This Mean for Developers?
The data presents a more complex picture than the binary of “for or against AI.” Use is expanding rapidly, but developers are reviewing the quality and limitations of outputs and continuing to seek autonomy and human context. In questions about a future in which systems perform most programming tasks, 61.3% said they would ask another person for help when they wanted to fully understand something, and 58.1% said they would do so to learn best practices.
From certi.news’s perspective, the review’s importance lies in the fact that it measures what is happening within work and learning, not merely the spread of tools. The actual change is AI’s transition from an optional experiment to a broad layer of work, alongside increased scrutiny of accuracy, understanding, and professional ownership. However, the source acknowledges that comparisons across years are incomplete because of differences in questions and samples, and that the relationship between AI use and improved learning remains an open question. The results of the 2026 survey will determine whether the gap between use and satisfaction persists and whether the adoption of agents will actually lead to the delegation of more decisions and tasks.