What Happened
JetBrains introduced Air in its 2026.3 IDEs as a workspace for running coding agents and reviewing their changes. It supports parallel sessions, Git worktrees, Code Notes, and IDE inspections, builds, and refactorings. It works with Junie and other agents through the Agent Client Protocol (ACP). JetBrains says 49% of developers spend at least four hours a week reviewing AI output and correcting prompts—a useful signal that review, not code generation alone, is the workflow to improve. [4]
GitHub added enterprise-level Copilot seat management for GitHub Apps. Apps can assign seats to users and enterprise teams when granted the new fine-grained Enterprise Copilot license management permission. Separately, personal-account repositories can now use customizable issue types and fields in project views, bulk updates, APIs, and webhooks. [5][2]
Why It Matters to Businesses
These changes address different parts of the AI development lifecycle. Air gives developers a place to inspect agent-produced changes using familiar IDE tools. GitHub’s seat API makes license assignment easier to automate, while structured issue data can give smaller teams a consistent way to describe and track work. None of these features replaces human approval or proves that generated code is correct. [4][5][2]
Kimbodo Engineering Perspective
The practical bottleneck is turning agent output into a reviewable change. Parallel sessions and worktrees help separate concurrent tasks, but they can also multiply changes awaiting review. We would use agents for bounded tasks, require an accountable reviewer for merges, and measure time spent reviewing and reworking changes—not just code produced. Air’s IDE inspections and builds are useful checks, not substitutes for project-specific tests and security review. [4]
How We Would Implement It
- Define a small agent workflow: one scoped issue, one worktree, a documented change summary, automated tests, and a human-reviewed pull request. Use Air’s Code Notes and IDE checks to make the reasoning and results easier to inspect. [4]
- Pilot with one approved agent and provider. Document which repositories it may access, what code may be sent to the provider, and who can authorize use. Air sends code only after a user chooses and authorizes an agent; the provider’s terms then govern its handling. [4]
- If automating Copilot licenses, give the enterprise GitHub App the license-management permission only where needed, reconcile assignments against approved teams, and audit changes. [5]
- For personal-account projects, use issue types and fields only where they improve triage or reporting; avoid creating a taxonomy that maintainers must continually administer. [2]
Risks, Costs and Security
Local Air workflows do not require a paid JetBrains AI subscription, but connected providers may impose charges and limits; some JetBrains AI features have separate requirements. Teams should check provider data terms before authorizing agents on sensitive code. Git worktrees separate working copies, not security permissions: agents and their tools still need constrained access to repositories, credentials, and execution environments. Enterprise Copilot seat automation also introduces a privileged App permission that should be limited and monitored. [4][5]
Where Kimbodo Comes In
Kimbodo builds and operates this in production for businesses — see our AI Application Development practice, or Estimate My AI Application.
