JetBrains Air: From the Developer Workbench to a System of Products for Agentic Development

JetBrains has introduced JetBrains Air — an open system of products for developers, teams, and organizations working with AI agents. It caps off six months of experiments: the company earlier launched JetBrains Central as a control system for agent-driven development, and now brings the pieces together.

The central claim is sobering: AI can produce code, but organizations still have to produce software. Agents have changed how software gets made, but they haven't changed what it costs to be wrong.

From the workbench to a system

For 26 years JetBrains focused on the individual developer's workbench. Now, the company says, the era when the whole development process fits in a single window is ending. The IDE stays important, but Air builds a system around it — for initiating, executing, coordinating, reviewing, and governing agentic work.

Air comprises: Air in JetBrains IDEs (directing agents and verifying their output), Air Teams (coordinating team workflows involving humans and autonomous agents), and Air Governance — the former JetBrains Central, handling policy, auditability, cost management, and accountability. The company's own coding agent, Junie, will be supported across all surfaces.

A multi-vendor bet

Openness is Air's foundational principle. JetBrains doesn't expect the market to consolidate soon: models excel at different things, rankings shift every few months, and teams within one company rightly make different choices. Through the Agent Client Protocol (ACP) and the ACP Registry, developers can plug in third-party agents without leaving the IDE.

The catch is what openness currently costs: every new tool strips away a little visibility, context doesn't carry between services, spend can't be attributed, and policies must be rebuilt for each one. Air offers a single place to run, see, and govern agentic development on top of any models and services.

The bottleneck is shifting

Individual adoption, the company notes, has outrun organizations' ability to build infrastructure around it. Obviously wrong code gets caught fast; the hard case is code that's almost right — plausible, passing a superficial check, yet quietly carrying a bad assumption. Code becomes cheaper to generate but more expensive to verify.

Work can be delegated, but accountability cannot: an agent won't get the 3 a.m. call when something breaks. That's why, JetBrains argues, control grows harder, not easier, as AI improves. The anchor is the deterministic code intelligence the company has accumulated over 26 years — it makes agents cheaper and more accurate without tying them to a single model.

Source: Lobsters