Link to sectionUse Cases
As expected in a developer survey, code generation ranked as the most common AI usage. On the other hand, even though image generation was the original use case for generative AI, only 38% of respondents stated using it.
Link to sectionAI Code Generation
Most of us aren't quite vibe coding just yet, with a majority of respondents (69%) generating less than 25% of their code through AI – and only a small minority (8%) generating more than 75% of it.
Link to sectionAI Code Refactoring
Even when AI is used to generate code, a large majority (76%) of developers stated they have to refactor at least half of the outputted code before it's ready to be used.
Link to sectionReasons for Refactoring
The top reasons for refactoring were cosmetic concerns such as poor readability, variable renaming, and excessive repetition.
Many respondents also used the freeform “other answer” field to state that generated code often just didn't work as intended.
Link to sectionCode Generation Frequency
This chart shows just how embedded AI has become in our daily workflows, with 46% of respondents using AI to generate code multiple times per day or more.
Link to sectionOther Tasks Frequency
Compared to code generation, AI is used for other tasks (research, summarization, translation, etc.) relatively less often – which makes sense given that coding is still what we spend the most time on.
Link to sectionGenerated Code
The most commonly generated code type proved to be helper functions, followed by frontend components, both of which are fairly self-contained, making them good candidates for code generation.
Many are also using AI to add documentation or comments to existing code, which is an unexpected use case.
Link to sectionPersonal Expenses
I'm not sure where AI companies are getting the cash to run their server farms, but one thing's for sure – it's not from individual developers, with a majority of respondents not currently spending any of their own money on AI tools and services.
Link to sectionCompany Expenses
Interestingly, company expenses follow a horseshoe pattern, with companies not spending anything on AI – unless they're spending over $5000! Whether this pricing model will prove sustainable for AI companies will remain to be seen.
Note that respondents also pointed out in freeform comments that they might not always have access to this information.
Link to sectionIndustry Sector
Don't hesitate to use our built-in Query Builder on any other chart to filter these survey results according to any specific industry sector.
Link to sectionLocal AI
Many respondents have already tried running their own AI models locally, despite the difficulties involved, and many others are interesting in trying.
This could become a key differentiator for new models, as this shows a strong demand for being more in control of your AI tools.
Link to sectionPain Points
Poor overall code quality ranked first when it comes to AI pain points.
Link to sectionMissing Features
The main thing missing from today's models is the ability to keep entire codebases in memory, something that will prove key if AI tools are meant to help us maintain applications, and not just prototype them.
Link to sectionHappiness
Despite the various pain points highlighted by the survey, respondents were overall quite positive on the state of AI for web development in 2025.