Among professional developers, the use and perception of AI are not so much a matter of seniority.
One could interpret Stack Overflow’s latest annual survey this way. It invites, in any case, consideration of the level of professional experience in many respects.
Illustration on the question « Do you use AI-based tools in your development process? ».
| 2024 | 2025 – overall | 2025 – 1 to 5 years (6360 respondents) | 2025 – 5 to 10 years (5997 respondents) | 2025 – more than 10 years (13,001 respondents) | |
| Yes | 63.2% | 80.8% | 73.6% | 69.6% | 64.5% |
| Daily usage | 50.6% | 55.5% | 52.8% | 47.3% | |
| Weekly usage | 17.4% | 18.1% | 16.8% | 17.2% | |
| Less frequent | 12.8% | 11.5% | 13.5% | 13% | |
| Envisaged | 13.5% | 4.6% | 2.5% | 3.7% | 6% |
| Not envisaged | 23.4% | 14.7% | 12.3% | 13.1% | 16.5% |
If we focus on professional developers (26,004 respondents), the rate of “yes” reaches 80.8%, compared to 63.2% last year. In contrast, only 14.7% now say they are not interested (vs 23.4%).
Daily use is somewhat more common among the younger profiles, who are also the ones least likely to rule out using such tools.
A form of prudence among the heaviest users
On certain points, Stack Overflow distinguishes between developers who use AI tools in a “majority” of their workflows and those who use them only “partially.”
This applies to the question “Which parts of your workflow do you currently perform with AI tools?”
| Task | 2024 | 2025 – majority (11,202 respondents) | 2025 – partial (20,991 respondents) |
| Code writing | 82% | 16.9% | 59% |
| Answer searching | 67.5% | 54.1% | 55.8% |
| Debugging | 56.7% | 20.7% | 47.1% |
| Code documentation | 40.1% | 30.8% | 30.3% |
| Generation of content or synthetic data | 34.8% | 35.8% | 28.6% |
| Learning codebase | 30.9% | 20.8% | 32.7% |
| Code testing | 27.2% | 17.9% | 27.5% |
| Commit/code review | 13.2% | 10.2% | 22.6% |
| Project planning | 12.2% | 10.8% | 17.1% |
| Predictive analytics | 5.3% | 11% | 12.7% |
| Deployment and monitoring | 4.5% | 6.2% | 10.5% |
| Learning concepts or technologies | 33.1% | 47.4% | |
| Creating or maintaining docs | 24.8% | 27.3% |
From year to year, the hierarchy of uses has changed little. The gap between the most mature and the least mature users has narrowed, especially when including the notion of “partial” usage.
Monitoring and predictive analytics, minority usage even among contemplated uses
Among those who do not use AI tools for the tasks in question, but who plan to do so, the situation is as follows:
| Task | 2024 | 2025 – majority planned (12,790 respondents) | 2025 – partial planned (22,518 respondents) |
| Code writing | 9.2% | 12.4% | 32.4% |
| Answer searching | 17.6% | 17.2% | 24% |
| Debugging | 25.9% | 14.8% | 30.9% |
| Code documentation | 38.2% | 28.6% | 30.5% |
| Generation of content or synthetic data | 33.1% | 28.9% | 28% |
| Learning codebase | 40.6% | 23.1% | 34.9% |
| Code testing | 46.2% | 25.8% | 34.7% |
| Commit/code review | 40.9% | 16.3% | 31.4% |
| Project planning | 31.7% | 14.3% | 24.8% |
| Predictive analytics | 39.8% | 23% | 25% |
| Deployment and monitoring | 39.6% | 15.1% | 25% |
| Learning concepts or technologies | 15.7% | 27.9% | |
| Creating or maintaining docs | 31.8% | 32.5% |
It should be noted that the time window has evolved from one survey to the next. In 2024, Stack Overflow looked at planned uses for the coming year. This time, respondents were asked to project over a 3- to 5-year horizon.
A growing reluctance to AI that grows without exception
There are 25,349 who do not intend to use AI for at least one of the tasks. Specifically:
| Task | 2024 | 2025 |
| Code writing | 5.9% | 28.9% |
| Answer searching | 8.1% | 19.6% |
| Debugging | 9.1% | 36.4% |
| Code documentation | 12.7% | 38.5% |
| Generation of content or synthetic data | 18.7% | 38.2% |
| Learning codebase | 18.7% | 39.4% |
| Test code | 17.1% | 44.1% |
| Commit/code review | 32.9% | 58.7% |
| Project planning | 42.3% | 69.2% |
| Predictive analytics | 38.7% | 65.6% |
| Deployment and monitoring | 40.6% | 75.8% |
| Learning concepts or technologies | 32.2% | |
| Creating or maintaining docs | 39.6% |
Reluctance grows without exception. They are twice as many to exclude using AI to generate content or synthetic data. Three times more for documenting code. Four times more for debugging. And almost five times more for writing.
A confidence not yet fully earned
Meanwhile, developers remain many in thinking that the AI tools they use handle complex tasks poorly
| 2024 | 2025 – overall | 2025 – 1 to 5 years (6258 respondents) | 2025 – 5 to 10 years (5922 respondents) | 2025 – more than 10 years (12,901 respondents) | |
| Very confident | 2.9% | 3.9% | 4% | 4% | 3.6% |
| Fairly confident | 31.5% | 25.2% | 28.1% | 25.4% | 23.5% |
| Neither confident nor skeptical | 20.9% | 14.2% | 13.4% | 13.8% | 14.9% |
| Not confident | 32.3% | 22.8% | 23.6% | 23.9% | 22.1% |
| Very unconfident | 12.5% | 18.6% | 19.2% | 19.5% | 17.9% |
As for problems and frustrations stemming from using AI tools, developers cited mainly in 2024:
- Lack of trust in the answers (66.2%)
- Inability to integrate the business context (63.3%)
- Inadequate policies to reduce security risks (31.5%)
- Lack of training on new tools (30.7%)
- Work generated by these tools (12.9%)
This year, the items have changed. The lack of trust remains pervasive (66% lament the “almost right” solutions). As well as the work generated (45.2% cite the time spent debugging generated code). Some also tend to lose confidence in their own ability to solve problems (20%). There are also difficulties in understanding how and why the generated code works (16.3%).
AI agents: preferred technologies
The 2025 survey introduced a section dedicated to AI agents. For those who use or develop them, respondents were asked whether they had used certain tools over the past year.
Storage (3398 respondents)
| Tool | Usage rate |
| Redis | 42.9% |
| GitHub Copilot Server | 42.8% |
| Supabase | 20.9% |
| ChromaDB | 19.7% |
| pgvector | 17.9% |
| Neo4j | 12.3% |
| Pinecone | 11.2% |
| Qdrant | 8.2% |
| Milvus | 5.2% |
| Fireproof | 5% |
| LangMem | 4.8% |
| Weaviate | 4.5% |
| LanceDB | 4.4% |
| Mem0 | 4% |
| Zep | 2.8% |
| Letta | 2.5% |
Orchestration (3758)
| Tool | Usage rate |
| Ollama | 51.1% |
| LangChain | 32.9% |
| LangGraph | 16.2% |
| Vertex AI | 15.1% |
| Amazon Bedrock Agents | 14.5% |
| OpenRouter | 13.4% |
| LlamaIndex | 13.3% |
| AutoGen (Microsoft) | 12% |
| Zapier | 11.8% |
| CrewAI | 7.5% |
| Semantic Kernel | 6% |
| watsonx.ai | 5.7% |
| Haystack | 4.4% |
| Smolagents | 3.7% |
| Agno | 3.4% |
| Phidata | 2.1% |
| Smol-AGI | 1.9% |
| Martian | 1.7% |
| Izyr | 1.5% |
Observability and security (2689)
| Tool | Usage rate |
| Grafana + Prometheus | 43% |
| Sentry | 31.8% |
| Snyk | 18.2% |
| New Relic | 13% |
| LangSmith | 12.5% |
| Honeycomb | 8.8% |
| Langfuse | 8.8% |
| Wiz | 6.9% |
| Galileo | 6.2% |
| ART (Adversarial Robustness Tookbox) | 5.5% |
| Protect AI | 5% |
| Vectra AI | 4.4% |
| Arize | 3.7% |
| Helicone | 3.2% |
| Metero | 2.7% |
| Opik | 2.3% |
Agents, copilots or assistants (8323)
| Tool | Usage rate |
| ChatGPT | 81.7% |
| GitHub Copilot | 67.9% |
| Gemini | 47.4% |
| Claude Code | 40.8% |
| Microsoft Copilot | 31.3% |
| Perplexity | 16.2% |
| v0.dev | 9.1% |
| Bolt.new | 6.5% |
| Lovable.dev | 5.7% |
| AgentGPT | 5% |
| Tabnine | 5% |
| Replit | 5% |
| Auto-GPT | 4.7% |
| Amazon Codewhisperer | 3.9% |
| Blackbox AI | 3.5% |
| Roo Code | 3.4% |
| Cody | 3% |
| Devin AI | 2.7% |
| Glean | 1.3% |
| OpenHands | 1% |
Agents perceived as beneficial but not revolutionary
Stack Overflow defines an AI agent as “an autonomous software entity that can operate with little to no human intervention using AI techniques.” Based on this definition, 14.9% of professional developers (base: 31,877 respondents) say they use them daily. 9.2% weekly. 7.7% less often.
17.2% plan to use them; 36.7% do not. An additional 14.2% of respondents say they use them “only as copilots or for semi-automatic data entry.”
Across tools, professional developers (base: 31,636 respondents) are 16.3% to say their work has been disrupted “to a great extent.” 35.3% feel it is “rather so.” 41.4% judge that not at all or only minimally.
If we look at a sample of AI agents users (12,823 responses), a majority agree on several benefits:
- Reduction in time spent on specific development tasks (70.1%)
- Increase in productivity (68.7%)
- More effective resolution of complex problems (64.2%)
- Faster learning of technologies and codebases (63.2%)
They are, however, fewer than half to agree on the following:
- Automation of repetitive tasks (49%)
- Improvement in code quality (37.5%)
- Enhancement of team collaboration (17.3%)
AI-generated illustration
