Web Developers and AI in 2025: What's Really Going On?

Is AI truly revolutionizing web development, or is it all just hype? A recent survey of over 4000 web developers reveals how they actually use AI tools. Let's dive into the surprising insights about AI code generation, preferred AI models, and the future of web development.

ChatGPT Dominates, But Developers Are Exploring Alternatives

ChatGPT is the most popular and well-loved AI model among web developers. 53.1% of developers have a positive view of it, versus only 7.3% with a negative view. Claude is another popular option. The average developer has experimented with nearly four different AI models, showing that they're always looking for new tools.

• Key takeaway: Don't get stuck on just one model. Explore to find the best fit for your projects.

The Biggest Hurdles: Hallucinations, Context, and Prompts

What's stopping developers from fully embracing AI?

• Inaccurate AI output: Developers don't trust AI models because of "hallucinations" or inaccuracies. If the output is unreliable, it doesn't matter how fast or cheap it is.
• Context limitations: AI struggles with large, existing codebases. They are much better with new projects.
• Poor prompt engineering: It's hard to know if AI output is wrong because of a bad prompt or the model's limitations.

Actionable Tip: Experiment with prompt engineering techniques to improve AI accuracy.

Coding Assistants Reign Supreme… For Now

Which AI tools are developers using right now?

• Coding Assistants: GitHub Copilot is the most popular option, adopted by 75% of respondents.
• AI IDEs: Cursor is the top choice, but only 42% of developers have tried an AI IDE.
• Code Generation Services: v0 leads this category, but only 31% of developers have used these types of tools.

The takeaway: coding assistants are preferred because they integrate seamlessly into existing workflows.

AI Code Generation: Common, But Refactoring is Essential

Interestingly, 91% of web developers use AI to generate code. But AI isn't taking over completely:

• Mixed usage: Only 13% of developers generate most (>50%) of their code using AI.
• Human oversight: Developers refactor an average of 61% of AI-produced code. Meaning, only 17% of all code written by AI doesn't need a human to refactor it.
• Refactoring reasons: The most common reasons for refactoring are poor readability, variable renaming, and excessive repetition.

Practical Advice: Understand that AI generates code that needs to be reviewed and refined by experienced developers.

Who Is Paying for AI? Maybe Not As Much As You Think

Are AI companies raking in the dough? Not necessarily.

• Individual Spending: 52% of developers don't pay for AI services personally.
• Company Spending: 39% of developers say their companies aren't spending anything on AI.
• Local Models: 78% are interested in running AI models locally to avoid subscription costs.

The Verdict: AI Is Here to Stay, But Proceed with Caution

Despite the issues, most developers don't want to go back to a pre-AI era. Here’s what the survey also showed:

• Integral Tool: 59% agree that AI tools are now an integral part of their workflow.
• Productivity Boost: The same percentage believe AI tools have made them "a lot more productive."
• Skill Concerns: 60% worry that relying on AI tools will lead to less skilled developers.

Final Thought: Use AI responsibly as a tool to enhance your skills, but don't let it replace your core coding expertise. Don’t jump in all at once, as horror stories of a faulty AI codebase and the human tasked to refactor it are around the corner.