Every year Stack Overflow surveys tens of thousands of developers worldwide. Three years of that data now tells a surprisingly nuanced story about AI and machine learning — one that's very different from the headlines.
Here's what the numbers actually say.
Adoption Is Climbing. Trust Is Not.
The headline stat everyone quotes is adoption:
- 2023: 70% of developers using or planning to use AI tools
- 2024: 76% — up 6 points
- 2025: 84% — up another 8 points
By that number alone, AI adoption in software development looks like a runaway train.
But look at the sentiment data running alongside it:
- 2023: 77% favorable or very favorable toward AI tools
- 2024: 72% — down 5 points
- 2025: 60% — down another 12 points
More developers are using AI every year. Fewer of them feel good about it every year. That gap is the most important story in three years of this data and almost nobody is talking about it.
The Trust Problem Is Structural
It's not that developers tried AI and had a bad experience. It's that the more they use it, the more clearly they see its limitations.
In 2024, only 43% of developers said they trust the accuracy of AI output. In 2025, 87% expressed concern about accuracy and 81% had concerns about security and privacy.
The most cited frustration in 2025? "AI solutions that are almost right, but not quite" — cited by 66% of developers. The second biggest? Debugging AI-generated code taking more time than writing it from scratch.
This is not a beginner problem. Almost half of professional developers in 2024 said AI tools struggle with complex tasks. Trust in AI answers is so low that 75% of Stack Overflow users said they'd still ask a human rather than trust AI output.
Machine Learning Is Now a Required Skill — Not Optional
Here's the shift that matters for anyone building a development career:
- 36% of developers learned to code specifically for AI in the last year
- 44% of developers used AI-enabled tools to learn new coding skills in 2025, up from 37% in 2024
- 67% of developers are learning to code for AI in the workplace or on personal projects
- Python usage jumped 7 percentage points from 2024 to 2025 — the survey specifically cites its role in AI, data science, and back-end development as the driver
The message is clear: understanding how machine learning systems work is no longer a specialisation. It's becoming table stakes for any developer who wants to stay relevant.
Who's Actually Using What
The LLM landscape among developers in 2025:
- OpenAI GPT models: 81% of developers
- Claude Sonnet: 43% (used more by professional developers than learners)
- Gemini Flash: 35%
For AI development tools and IDEs:
- Cursor: 18% adoption
- Claude Code: 10%
- Windsurf: 5%
ChatGPT had a 75% admiration rate in 2024 — meaning most developers who used it wanted to keep using it. That's a strong signal. But it also means the market is still being shaped, and alternatives are gaining ground fast.
AI Agents: Promising But Overhyped
The 2025 survey asked directly about AI agents — autonomous AI systems that complete multi-step tasks without human intervention.
The verdict is not what the hype suggests:
- Only 31% of developers currently use AI agents
- 17% plan to use them
- 38% have no plans to use them at all
- Only 17% of agent users say agents have improved team collaboration
The individual productivity gains are real — 69% of agent users report increased personal productivity. But "this made me faster" and "this transformed how teams work" are very different claims. The data supports the first. Not yet the second.
What This Means for Learning Machine Learning
Three conclusions stand out from the three-year trend:
1. Learning ML fundamentals is more valuable than learning specific tools
Tools are shifting fast — the top AI IDE in 2025 barely existed in 2024. But the underlying concepts — how models are trained, what bias-variance trade-off means, why overfitting happens — those don't change. Invest in the fundamentals.
2. Python is the language of AI, full stop
The 7-point surge in Python adoption is directly tied to AI. If you're choosing a language to learn for ML/AI work, this is an easy call.
3. The trust gap is a career opportunity
If 66% of developers are frustrated by AI that is "almost right but not quite" — someone needs to understand these systems well enough to catch and fix those errors. That's not a job AI can do. It's a job for a developer who deeply understands how ML models work.
The adoption curve is steep. The understanding curve is lagging behind. That gap is exactly where skilled developers will differentiate themselves over the next few years.
Data sourced from Stack Overflow Developer Surveys 2023, 2024, and 2025. If you're building ML fundamentals to navigate this landscape, thecodeforge.io/ml-ai/ covers the core concepts worth knowing.
Top comments (0)