JetBrains Central — Governance Infrastructure for the Agent Era

On March 24, 2026, JetBrains announced Central — a governance and execution platform for multi-agent software development. The same week, they published the Code With Me sunset notice: the 2026.1 IDE release is the last to ship with it built in, plugin support runs until Q1 2027, then the relay infrastructure goes dark. These two moves are not coincidental. JetBrains is explicitly closing the chapter on human-to-human collaboration tooling and opening one on human-to-agent orchestration.

TL;DR

• What: JetBrains Central is a governance and execution platform for AI coding agents — policy enforcement, cost attribution, agent runtimes, and shared semantic context across repos
• Timing: EAP launches Q2 2026 with design partners; production deployments are realistically six months away
• The bet: JetBrains is retiring Code With Me and going all-in on agent infrastructure — not incremental, a directional commitment
• Action: Watch it, don’t buy it yet — the problem it solves is real, the product isn’t ready

JetBrains Central — What Happened

I’ve watched JetBrains survive three platform shifts by being the “professional” option — the choice for teams that need real tooling, not toys. Central is that same bet applied to the agent era. Not raw autocomplete, not a flashy new editor. Governance, cost attribution, and policy enforcement for teams running multiple agents across multiple models. The unsexy infrastructure play that enterprise teams will actually need once they get past the “demo impressive, production broken” phase of agentic AI.

The data JetBrains is working from is worth taking seriously. Their January 2026 AI Pulse survey of 11,000 developers found 90% already use AI at work, but only 13% use it across the full software development lifecycle — code review, release pipelines, the whole chain. Twenty-two percent are using AI coding agents today. Sixty-six percent of companies plan to adopt them within 12 months. The gap between those numbers is the market JetBrains is building for: the teams that have proven agents work in isolation and are now staring at the question of how to run them at scale without losing control of cost, security, or output quality.

Why This Matters

Central is three stacked layers, and the distinction between them matters for evaluating whether this is real infrastructure or a marketing reframe.

The first layer is governance and control: IAM, policy enforcement, and cost attribution per agent. This is the piece that doesn’t exist anywhere well today. You can wire up Claude or Codex to your CI pipeline, but attributing spend to specific delivery outcomes — which agent ran what task, how much it cost, what it produced — is manual accounting at best. Central makes this the product.

The second layer is agent execution infrastructure: cloud runtimes with sandboxing and failure recovery. This is closer to commodity territory. Managed compute for agentic workloads is something cloud providers are building toward, and the moat here is thinner.

The third layer is the one that could be genuinely differentiated: a shared semantic context layer that aggregates knowledge from codebases, architecture, runtime behavior, and delivery infrastructure across repos and projects. The idea is that agents operating through Central get system-level understanding, not just file-level context. If JetBrains actually delivers this — and it’s a meaningful if — it’s the kind of thing that’s hard to replicate without 26 years of IDE data and language tooling investment.

Central is LLM-agnostic by design. It orchestrates JetBrains’ own Junie agent alongside OpenAI’s Codex, Anthropic’s Claude Agent, and Google’s Gemini CLI. The Junie CLI, which hit beta on March 9, is explicitly built to be model-agnostic. Launch partners are Google Cloud, Anthropic, and OpenAI — the same players who benefit from governance infrastructure that makes enterprise adoption of their models less risky for procurement teams.

The LLM-agnostic architecture matters for enterprise buyers specifically. Central’s value proposition isn’t tied to which model wins — it’s the layer that makes switching models a configuration change rather than a migration project. That’s a real procurement argument.

The pricing model follows the logic of the platform. Fixed per-seat subscription for the governance layer, pay-as-you-go for agentic execution. Oleg Koverznev, VP and Head of Agentic Platform at JetBrains, framed the range directly: “One developer can spend $100 a month. Another can orchestrate thousands of agents and spend $100,000.” The pitch is cost legibility — connecting agent spend to actual delivery outcomes rather than treating AI as an undifferentiated line item.

Compare this to how Coder approached the governance problem after its $90M Series C: they’re coming from the infrastructure side, building secure compute environments for agents to run in. Central is coming from the developer tooling side, building the orchestration and context layer above the compute. These aren’t the same product, and the enterprise teams that will care about governance will likely need both — which is worth tracking as the space matures.

The agentic infrastructure stack is still being assembled in real time. Central is one piece of a picture that doesn’t have clear standards yet. The bet JetBrains is making is that the governance and context layers will be as important as the execution layer — and that they’re better positioned to own those than cloud providers or the model vendors themselves.

Central is pre-EAP. The Early Access Program launches Q2 2026 with a limited group of design partners. Real production deployments are six months away at minimum. If you’re evaluating agent governance infrastructure for a 2026 rollout, this is a watch item, not a procurement decision.

The Air/Central integration is the part I’d watch most carefully. JetBrains Air — currently macOS-only, with Windows and Linux planned for 2026 — gives individual developers a workspace for running parallel agent tasks. Central adds the organizational governance layer on top. Whether these two products feel like a coherent system or two separate products sharing a brand will define whether JetBrains has actually built something end-to-end or assembled a portfolio of bets. The integration between them isn’t fully documented yet, and that gap is where enterprise evaluation will live.

The Code With Me shutdown is a signal worth reading separately from the Central announcement. Code With Me was never a flagship feature, but unbundling it from the IDE entirely and putting it on a plugin sunset timeline is a resource allocation statement. JetBrains is not building for human pair programming. They’re building for human-agent workflows. That’s a bet that the collaboration patterns of the next five years look fundamentally different from the last five.

Code With Me will be available as a standalone plugin via JetBrains Marketplace through Q1 2027. After that, the public relay infrastructure shuts down entirely. Teams using it for remote collaboration have until Q1 2027 to migrate.

The Take

JetBrains Central is the right product for the right moment, arriving about six months too early to evaluate seriously. The problem it’s solving — governance, cost attribution, and semantic context at scale for multi-agent teams — is real and underserved. The data from their own survey makes the market case clearly. And JetBrains has the IDE integration depth and language tooling history to build a semantic context layer that pure-play orchestration tools can’t easily replicate.

But the EAP doesn’t start until Q2 2026, and EAP in practice means limited access, rough edges, and no production SLAs. If you’re a tech lead evaluating agent governance infrastructure right now, the honest timeline is: nothing to deploy before Q3 2026 at the earliest, and enterprise-grade reliability probably means 2027.

Watch the Air/Central integration — that’s where the coherence of the platform will be tested. Watch whether the semantic context layer actually delivers cross-repo intelligence or turns out to be a better-marketed version of what existing code search tools already do. And watch whether JetBrains can execute on a platform play while simultaneously shipping the IDE releases their existing customer base depends on.

The directional commitment is real. JetBrains isn’t hedging — killing Code With Me while launching Central is a statement, not a roadmap item. Whether the execution matches the ambition is what the next six months will tell you.