Before the results of the 2026 Developer Survey are released, Stack Overflow is revisiting data from 2024 and 2025 to track broader changes in how developers use AI, organize work, learn, and rely on technology communities. The clearest picture is not unconditional adoption of new tools, but expanding use alongside declining enthusiasm and greater scrutiny of output quality.
From Coding Assistants to Agents
The share of developers who said they use AI tools rose from 44% in 2023 to 62% in 2024 and then 79% in 2025, while nearly half of respondents reported using them daily. In the 2025 survey, ChatGPT had the strongest presence among users of ready-made assistants at 82%, followed by GitHub Copilot at 68%, Google Gemini at 47%, Claude Code at 41%, and Microsoft Copilot at 31%.
As for AI agents, their use reached 31% in the 2025 survey before rising to 59% in a smaller pulse survey conducted in April 2026. During the same period, Claude Code's share rose from 41% to 55% in the question specifically concerning agent use. The material indicates that use was higher among daily users and people working in execution-oriented roles, but these figures do not mean that agents took over most engineering decisions.
Tool-use data shows that coding, research, and debugging remained major areas, although direct comparison between 2024 and 2025 is limited because of changes in question design. Stack Overflow's interpretation is that automation is concentrated more in discovery and drafting, while actual engineering work remains largely AI-assisted rather than independent of AI.
Learners Use the Tools but Trust Them Less
The share of students and programming learners who use AI resources rose from 63.0% in 2024 to 73.3% in 2025, but positive attitudes fell from 72.2% to 52.8%, while skeptical attitudes rose from 6.4% to 26.3%. Medium or high confidence in output accuracy also declined from 49.1% to 37.4%.
These findings align with practical difficulties reported by participants: 20.0% said that using AI made them less confident in their ability to solve problems, and 16.3% said they found it difficult to understand how code works or why it works. These figures do not prove that the tools weaken learning, but they distinguish between producing code that works and understanding it well enough to explain and modify it.
Flexible Work and Professional Satisfaction
The share of people working fully remotely fell from 38.0% in 2024 to 32.4% in 2025, while 17.9% reported working entirely on-site, 37.1% reported structured hybrid models, and 12.6% reported high flexibility in which attendance is determined according to personal preferences or project needs. Counting models that provide some degree of flexibility, the share reached 82% in 2025 compared with 80% in 2024.
The picture varies geographically: the United States recorded the highest rate of fully remote work at 45.0%, while India had the highest share of on-site work at 30.5%. Germany recorded 20.7% in the flexibility category in which employees choose whether to attend, and 24.6% in the flexible hybrid model, while the two figures in the United Kingdom were 14.0% and 25.8%.
At the satisfaction level, 28.4% of participants described themselves as unhappy, 47.1% as being in a state of mild satisfaction or habituation, and 24.5% as genuinely happy. The share of those who see AI as a threat to their jobs also rose from 12.1% to 15.0%. The 2025 results place autonomy and confidence in managing tasks at the top of satisfaction factors, ahead of pay, benefits, and solving real-world problems.
What Does This Mean for Developers?
The data shows that AI-powered tools are entering the existing work stack rather than immediately displacing it. In enterprise environments, Cursor recorded 17.9%, Claude Code 9.7%, Windsurf 4.9%, and Aider 1.9%, alongside GitLab's presence at 35.6%. Docker's position also rose to 71.1% in 2025, Kubernetes to 28.5%, and PostgreSQL to 55.6%.
The need for human context remains clear: in 2025, 61.3% of participants said they would ask another person for help if they wanted to understand something fully, while 58.1% said they would do so to learn best practices. Accordingly, the open-ended questions in the 2026 survey focus on whether increased AI use is accompanied by greater delegation, how control over work affects satisfaction, and whether the tools support understanding rather than merely completing the task.