An AI IDE is an integrated development environment in which AI is a core architectural component rather than an optional extension. The distinction Rule Taxonomy and Evolution in AI IDEs: A Mining and Survey Study draws is against the plugin arrangement — VS Code with a GitHub Copilot extension, for instance — where the assistant operates within the constraints the host editor imposes, limiting its access to runtime state, internal editor signals and fine-grained interaction data. In an AI IDE the model is integrated into the core editor infrastructure instead, which that study argues is what enables capabilities beyond code completion: autonomous task decomposition, multi-file refactoring, architectural reasoning, context-aware generation, and continuous alignment with project-specific configuration.
Usage
The category as that study uses it is defined partly by a shared feature: the tools it examines all let developers define AI IDE Rules that the IDE must follow during code generation and chat. The five it studies on that basis are Cursor, Windsurf, Trae, Qoder and Kiro.
That study’s mining of 83 projects declared as built with an AI IDE gives a picture of where the category was being used at the time. TypeScript dominates the codebases at 50.6%, followed by Python at 14.4% and JavaScript at 9.6%; web application development accounts for the majority of domains, with full-stack, frontend and backend projects together making up 56.6%. The projects are small: over half have fewer than 50 commits, and solo developers or teams of two to three account for the large majority. The study reports that many of them were created between June 2025 and September 2025.
The survey side gives a different view of the same question, since respondents report on their own practice rather than on a filtered project set. Among 99 practitioners, Cursor was the most widely adopted at 57, closely followed by the VS Code plus Copilot setup at 51 — that is, a plugin arrangement rather than an AI IDE — with Kiro at 37, Trae at 24, Windsurf at 23 and Qoder at 16. A further 34 reported using Antigravity, and Claude Code, Codex and OpenCode were named under an “Other” option.
When It Applies
The term is useful where the architectural distinction it names actually makes a difference — that is, where the depth of the model’s access to editor state and project context is the thing under discussion. Where it is not, “AI coding tool” covers the same ground with fewer commitments; the survey data above is the clearest illustration, since a plugin setup sits second in adoption among developers who also use AI IDEs, and the study counts it separately without treating it as one.
The study offers no operational test for membership beyond the architectural characterisation and its own selection of five tools that support user-defined rules, so where a given tool falls is left to judgement. It does note that the ecosystem is evolving rapidly — both the underlying foundation models and the rule-injection mechanisms undergo frequent updates — and presents its own findings as a snapshot of current context-engineering practices.
Related Terms
- AI IDE Rules — the configuration mechanism these tools share
- Rule Taxonomy and Evolution in AI IDEs: A Mining and Survey Study — the study this account is drawn from
- AI Coding Agent — the agent-shaped sibling of this category
- Agentic Coding — the practice these tools are built for