IntelliJ IDEA 2026.3 Ships Agents Into the IDE, Makes Debugging Visible, and Settles Long-Neglected UI Debts
This release turns agent use from a bolted-on plugin into a first-class IDE workflow, letting teams keep their existing agent subscriptions while getting a structured review loop. The auto-reporting of error and UI freeze telemetry changes the privacy calculus for enterprises, since it ships on by default and must be explicitly disabled.
The biggest structural change is Air becoming a first-class, bundled agent development layer. It detects existing ACP-compatible agents like Claude Code and Codex, reuses your current provider subscriptions and API keys, and lets you anchor review comments directly on diff lines to send feedback back to the agent. Skills, MCP configs, and chat history migrate from AI Assistant automatically.
Debugging gets a triple boost: return values now appear in the Variables view at method exits; the Hibernate Debugger plugin shows generated SQL, bound parameters, HQL/JPQL, and source lines for repository calls; and the ij-debugger skill gives agents access to log points and runtime diagnostics that persist past the debug session. Spring Security failures inject an explanatory header into the HTTP response to show which rule or permission rejected a request.
Everyday improvements include a Home window that opens a blank file or terminal without loading a project, built-in SSH remote development with Linux Light mode, a terminal on libghostty compatible with Neovim and AI agents, and sound notifications for build, test, and agent events. Database support adds a DuckDB-backed Workspace data source for cross-database joins from one SQL console. JetBrains reports over 1000 bug fixes and 200+ performance fixes, and from 2026.3 it automatically sends sanitized error and UI freezes reports unless disabled.
JetBrains is separating agent orchestration from AI chat, which means the IDE treats agents as a workflow tool with its own review cycle rather than a chat widget with code generation.
The decision to not lock agents to JetBrains AI and instead adopt ACP is a bet that agent clients are a commodity; the IDE's differentiator becomes the management and review surface it provides.
Auto-reporting error and UI freezes with sanitization is a quiet trade-off: users get faster fixes, but the default-on telemetry will likely trigger scrutiny in privacy-sensitive shops.
Debugger improvements like return values and Hibernate tracing target the daily friction of enterprise Java work, signaling that JetBrains still prioritizes traditional server-side development over AI-only features.
The libghostty terminal collaboration points to an ecosystem where terminal emulation becomes a shared library, letting JetBrains keep up with modern terminal protocol demands without reinventing them.