[1/2] State of Web Dev AI 2025 — Results for Engineers (Category AI)
Almost a year ago, Devographics ran its first developer survey on the state of AI in web development. I read the report back in the autumn but somehow forgot to write about it, so here it is.
Report methodology The survey ran from February 10 to March 10, 2025, with 4,181 web developers taking part. It explored how developers use AI, which tools they find most useful, and the problems they encounter. Participation was open, so the sample probably contains a disproportionate number of AI enthusiasts; recruitment included State of JS/CSS subscribers. Here are the findings.
How developers use AI 🤖 Code generation is the main use case: approximately 82% of respondents use AI to write code. By comparison, only 38% use image generation. Despite the attention around Midjourney and similar products, visual tools remain a niche use case in web development. 😑 AI still produces a relatively small share of the code. For 69% of respondents, AI generates less than a quarter of the final code; only 8% use it to produce more than 75%. For most developers, assistants write individual fragments rather than entire projects. ⚙️ Frequency and effectiveness: nearly half of developers (46%) generate code with AI several times a day or more. Many have incorporated these tools into their daily workflow. Most respondents also agree that AI assistants have substantially improved their productivity.
Popular AI tools 🤖 OpenAI's ChatGPT leads in reach: 91% of web developers have at least tried it at work. Other major models are attracting users too: roughly 55–60% of respondents have experimented with Anthropic Claude, Microsoft Copilot (the model backend), or Google Gemini. The figure for the newer xAI Grok is approximately 25%. 🧑💻 GitHub Copilot is the most widely used coding assistant, an AI code-completion plugin. Around 71% of respondents use it, and it leads in positive feedback. Other assistants, such as Tabnine and JetBrains AI, lag far behind at roughly 10–15%. Supermaven stands out: fewer than 10% have tried it, but the feedback is very positive—a possible dark horse to watch.
Problem areas ⚠️ Code reliability and quality. Hallucinations, factual errors, and limited model context are the main barriers to broader adoption. As a result, 76% of developers rewrite at least half of the AI-generated code before using it. Often, the fragment simply does not work as intended. AI code still needs careful review and testing. 💰 Tool costs. Most developers use free AI services: more than 90% spend under $50 a month, including 52% who pay nothing. Companies are cautious too: approximately 38% of teams make no AI investment, while around 12% already spend over $5000 a month. Many are still assessing the actual return from these tools.
The next post covers survey findings that may interest technical leaders.
#Engineering #AI #Metrics #Software #DevEx #Productivity #DevOps