Chuyện gì đã xảy ra
36Kr reports that JetBrains Research’s 2026 Developer Ecosystem Survey found rapid growth in AI coding-agent use among more than 15,000 professional developers worldwide. About 90% reportedly used an AI programming agent at work at least once a week, while 68% used one daily. The survey included software developers and engineers as well as AI/ML, DevOps, architecture, data, and quality-assurance roles.
36Kr reports that JetBrains Research surveyed more than 15,000 professional developers worldwide for its 2026 Developer Ecosystem Survey, with roughly 90% of respondents described as developers, programmers, or software engineers. The reported survey period was May through July 2026. The source says the survey also covered AI/ML engineers, DevOps engineers, architects, data engineers, and QA engineers, so its headline results describe a broad professional technology audience rather than only conventional application programmers.
According to 36Kr, about 90% of professional developers used AI programming agents at work at least once a week, and 68% used them every day. JetBrains’ reported definition of an agent included locally running and cloud-based agents. The source presents those figures as evidence that agents are moving beyond occasional code completion or question answering toward participation in larger development tasks.
36Kr reports that Claude Code’s workplace adoption rose from about 18% in January 2026 to 39% in May–July. In the United States, the reported figure was 47%. The share naming Claude Code as the primary or most commonly used AI coding tool was reported at 31%. GitHub Copilot’s workplace adoption was reported at 21% in May–July, down from roughly 29%–31% between mid-2025 and early 2026. The source says both products had 79% awareness. It also reports OpenAI Codex adoption rising from 3% to 16%, Cursor declining from 18% to 12%, open-source OpenCode reaching 7%, Google Antigravity reaching 6%, and JetBrains’ own tools reaching about 9%.
Tại sao nó quan trọng
The figures suggest that AI coding tools are becoming part of routine software work, although the survey does not establish how much productive work agents complete or how reliable their output is. 36Kr’s account also describes a shift in market competition from editor-based assistance toward tools that can handle broader development tasks, including codebase changes, testing, and repeated execution.
The reported ranking matters because it indicates that the market for AI-assisted software development may be fragmenting around different forms of assistance. 36Kr describes Claude Code and Codex as agents that can operate beyond a traditional integrated development environment, while Cursor’s reported decline is presented alongside continuing awareness growth. If the pattern is representative, recognition of a tool and repeated workplace use are becoming distinct measures of competitive strength.
The reported weekly and daily usage rates also point to a possible change in how software teams allocate work. Agents may be used for repository navigation, code modification, test execution, debugging, and other multi-step tasks rather than only for generating snippets. That could affect developer workflows, review practices, onboarding, and demand for tools that manage permissions, testing, version control, and audit trails. The source does not show, however, that agents independently complete these tasks successfully or that they reduce development time.
For buyers and workers, adoption figures alone are an incomplete guide. The survey as described does not provide error rates, security incidents, code-quality measurements, total cost, task mix, or evidence that one product produces better outcomes than another. It also does not establish whether respondents could select multiple tools, how “use” was measured, how the sample was recruited or weighted, or whether the results generalize to smaller firms, nontechnical workplaces, or developers outside the surveyed population. Those limits make the figures useful as an adoption snapshot, not as a productivity verdict.
Cơ chế tương tác: Nó thực sự hoạt động như thế nào
Khám phá công nghệ cơ bản đằng sau sự phát triển này một cách tương tác.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Xem gì tiếp theo
The main questions are whether Claude Code can sustain its reported lead, whether OpenAI Codex continues gaining adoption, and whether GitHub Copilot or Cursor reverse their reported declines. More useful evidence would include methodology, sample weighting, task-level productivity, error rates, costs, security controls, and independent comparisons of the tools’ real-world performance.
The next development to watch is whether the reported gap persists in later surveys. Claude Code’s move from 18% to 39% in the source’s figures is substantial, but a single survey period cannot show whether the increase reflects durable adoption, temporary experimentation, changes in question wording, or differences in the respondent population. A follow-up using the same questions and sampling method would make the trend easier to interpret.
Codex, GitHub Copilot, and Cursor will be important comparison points. 36Kr reports strong growth for Codex, a decline for Copilot, and a decline for Cursor despite higher awareness of Cursor. Future reporting should distinguish workplace availability from primary-tool status, frequency of use, paid usage, and use for consequential production code. Those distinctions could change the meaning of the current rankings.
The most consequential evidence would come from deployments and evaluations that measure completed tasks rather than stated use. Useful indicators would include independently audited success rates, review burden, regressions, testing performance, security and privacy protections, permission controls, , and cost per completed task. The source does not report these measures, and it does not independently confirm JetBrains Research’s underlying survey results. Until such evidence is available, the clearest supported conclusion is that 36Kr reports broad and growing use of AI coding agents, alongside a reported reshuffling of tool adoption. Taken together, the figures describe reported workplace adoption and awareness, not a definitive league table of technical capability. Their value is greatest when read alongside the survey scope, the stated time period, and the methodological questions that remain unanswered in the available account.