[1/2] State of Web Dev AI 2025 Analysis of results for engineers (AI column)
Almost a year ago, the Devographics platform conducted the first developer survey to assess the state of AI for web development. This one report I studied it in the fall, but somehow forgot to write about it, so I will correct it now.
Report methodology The survey took place with 10 February 10 March 2025 Years of participation 4 181 Web developer. The questionnaire studied how programmers use AI in their work, which tools are most useful and what problems they face. The survey was open to everyone, so there are probably more AI enthusiasts in the sample. (Audience recruited from State of JS/CSS subscribers). As a result, such results were obtained.
How developers use AI 🤖 Code generation – the main application scenario: ~82Percent of respondents use AI to write code. For comparison, the generation of images uses only 38% – Despite the buzz around Midjourney and others, visual tools remain niche in web development. 😑 The share of AI code is still small. U 69% of respondents generate less than a quarter of the final code, and only 8Percentages are obtained using AI more 75Percent of its code. In other words, for most developers, assistants are still writing fragments, not the entire project. Frequency and effectiveness: nearly half (46%) Programmers run code generation using AI several times a day or more often. Many have included these tools in their daily workflow. Moreover, most respondents agree that AI assistants have significantly increased their productivity.
Popular AI Tools ChatGPT from OpenAI is the absolute leader in coverage: 91Some web developers have at least tried it. Other major models are also gaining users:55–60Percent of respondents experimented with Anthropic Claude, Microsoft Copilot (model-back) Or Google Gemini. For comparison, the new xAI Grok has only ~25%. GitHub Copilot is the most popular coding assistant (AI plugin for autocomplement code). It's being used.71% of respondents, and he leads in the share of positive reviews. For comparison, other assistants like Tabnine or JetBrains AI are far behind. (order 10–15Percentage of users). But Supermaven stands out: they have tried it so far.10%, but the reviews are very high – a possible “dark horse” that can shoot in the near future.
Problem points The reliability and quality of the code. Hallucinations, factual errors, and the limited context of the model are major obstacles to the widespread use of AI tools. Eventually. 76Percentage of developers rewrite at least half of the code generated by AI before using it. Often the reason is simple: the resulting fragment does not work as intended. Without a thorough review and tests, you cannot trust the AI code yet. ы Tool costs. Most developers use free versions of AI services: more 90Percent spend less than $50 monthly, including 52You don't pay anything. Companies are also cautious about budgeting:38The percentage of teams don’t invest in AI at all.12% are already spending large amounts (> $5000 monthly). While many are waiting, assessing the real impact of AI tools.
In continuation A story about survey results that may be of interest to tech leaders
#Engineering #AI #Metrics #Software #DevEx #Productivity #DevOps