How AI Is Redefining IDEs: Inside JetBrains AIR’s Native AI Architecture

AIR: What Is It?

JetBrains AIR is a brand‑new, standalone IDE built from the ground up with AI as the default workflow rather than an optional feature. Its core principles are:

AI is the default workflow, not a toggle.

Emphasis on determinism and explainability to avoid "black‑box" magic.

Developers retain full control over the generated code.

Quick Start: Three Steps to Experience AIR’s AI‑Native Flow

1. Install and First Run

Currently available only for macOS (Windows/Linux planned later). After installing, log in with a valid JetBrains subscription and you are taken to the welcome page.

2. Open a Project

You can either open a local folder (click Open ) or clone a Git repository (click Clone from Git , provide the repo URL and destination). On first opening, AIR asks whether to trust the code because executing code carries risk; you can choose Preview (limited, no execution) or Trust (full execution, recommended only for trusted sources).

3. Define a Task

Interact with the AI agent via a chat‑like interface. Provide a clear task description, e.g., “Add pagination to the order module, 10 items per page,” and optionally supply context such as a file path (e.g., src/orders/service.py). AIR also supports a Plan Mode where the AI first generates an execution plan before refining the task.

Task Configuration and Permission Modes

Within the task settings you can adjust:

Permission prompts : AI asks for permission the first time a tool is used.

Auto‑edit : Automatically accept file edits.

Planning mode : Only analyze code without editing or executing.

Full access : Skip all permission dialogs.

Context can be added from files/folders, Git branches/commits/local changes, MCP servers, terminal tabs, or by uploading local files. You can also select a code snippet directly and add it to the task.

Running and Managing Tasks

Task states include:

Running : AI is executing; you can supplement input at any time.

Waiting for user action : AI pauses awaiting manual intervention.

Finished : Task completed and ready for review.

Canceled / Archived : Manually stopped or archived after exit.

Multiple tasks can run in parallel (e.g., one generating tests while another fixes a bug) without interference.

Review and Commit Changes

In the Review tab, a diff panel shows AI‑made modifications. Two diff views are available:

Unified diff : Single‑page side‑by‑side comparison.

Split diff : Left‑right column comparison.

You can add comments next to line numbers, just like a code review, to provide feedback.

Advanced Feature: MCP Server Integration

Model Context Protocol (MCP) lets AIR connect to external tools such as databases or APIs. Configuration is done by pasting a JSON block into Settings → AI → MCP Servers:

{
  "mcpServers": {
    "alphavantage": {
      "command": "uvx",
      "args": ["av-mcp", "YOUR_API_KEY"]
    }
  }
}

This example connects to the Alpha Vantage API.

Summary of the Workflow

Define the task with contextual information and incremental guidance.

Configure execution environment and permissions, balancing efficiency and safety.

Review and commit changes to ensure controlled code quality.

Combined with parallel task execution, MCP extensions, and local snapshots, AIR delivers a more efficient and controllable AI‑assisted development experience for complex projects.