2026年代码状态开发者调查报告

Table of Contents Introduction03 About this report04 How developers are (really) using AI05 Vibe check: Do developers trust AI?09 The second act of AI: Agents20 Introduction Sonar analyzes over 750 billion lines of code each day, which gives us a uniqueunderstanding of code. This year, we kicked off a new report series called the State of specific coding personalities of leading LLMs. Those reports focused primarily on thecode itself and the models creating it. Next we wanted to expand that view to include thestate of code from the perspective of the people doing the work—the developers writing Specifically, we wanted to get a read on what is changing for them. As AI rapidlyshifts the mechanics of coding, we need to understand the on-the-ground reality—theefficiencies, the frustrations, and the new workflows emerging. To ensure we add real After surveying more than 1,100 developers, we saw a critical new narrative emerging.Simply put, the explosion in AI-generated code hasn't led directly to massive and much-hyped productivity gains yet. Instead, a verification bottleneck has emerged, creating About this report The 2026 State of Code Developer Survey was a quantitative online survey conductedamong professional software developers. Fieldwork for the survey ran throughout The final sample size for the study included 1,149 responses, distributed globally. Allrespondents were 18 years or older, employed full-time or self-employed in a technologyrole (the vast majority worked in software engineering, with some others in fields related Further details about the report’s demographic makeup are available in the appendix. How developers are (really) using AI 72% of developers who have tried AI use it every day AI-assisted coding is officially a standard part of the developer workflow. 72% of developers who have 72% of developers who have tried AI use it every day How frequently (do you / your team or company) use AI coding tools in your development workflow? Developers also report that 42% of their code is currently AI-generated or assisted—a share that theypredict will increase by over half by 2027, and up from only 6% in 2023. Average share of AI-assisted or generated code committed by developers What % of the code you committed or contributed was / will be generated or significantly assisted by AI tools? And AI is not just for side projects and experimentation. Developers are using AI across the gamut ofsoftware projects, from prototypes (88%) and internal, non-critical production software (83%) all the Developers are using AI across the gamut of software projects Thinking about your team / company, which of the following types of work involves the use or assistance of AI? Use cases, and the gap between usage and effectiveness Just because AI is used everywhere doesn't mean it's effective evenly. When we look at howdevelopers are using AI versus how effective they find it for those specific tasks, a clear gap sometimesemerges. In a perfect world, adoption would increase more or less linearly with effectiveness. But Understanding AI use cases For which of the following tasks is your team /company using AI coding tools? How effective are AI coding tools for each ofthe following tasks you or your team / company The best example of this is also the most common use case for AI: assisting with new code development(90% of developers). Only 55% of those users rated AI as "extremely or very effective" for that task. Refactoring or optimizing existing code shows a similar effectiveness gap: while 72% of developersreport using AI tools for this use case, only 43% attest to its effectiveness in that task. Where AI really shines The data shows AI tools are most effective at tasks that involve experimentation or working with what's The tasks where AI is most effective include: •Writing documentation (74% effective) •Explaining or understanding existing code (66% effective) •Vibe coding / green-field prototyping (62% effective) •Generating tests (59% effective) Developers have embraced AI as a daily partner, but they're finding it's a much better "explainer" and"prototyper" than it is a "maintainer" or "refactorer"—at least for now. It's highly effective at generatingnew things (docs, tests, new projects) but struggles more with the complex, nuanced work of modifying The takeaway Developers are pragmatic: they’ve fully embraced AI as a daily assistant, using it to writedocumentation and generate tests. But they also know its limits, showing less confidence in itsability to handle complex, existing code. This gap between high usage and selective effectiveness Vibe check: Do developers trust AI? 96% of developers don’t fully trust that AI-generated code isfunctionally correct It's no secret that AI is changing development. Our study found that developers are seeing real benefits,reporting an average personal productivity boost of 35%. On top of that, more than half (54%) say And while