Visual Studio Code 1.124 Gets Another Round of Agent-First Refinements

The June 7-8 updates to Visual Studio Code 1.124 deliver a focused batch of improvements that make the agent-centric development experience significantly more reliable, navigable, and enterprise-ready. While 1.124’s initial June 5 release brought WSL agent support and background send, this mid-cycle update goes deeper — adding session layout persistence across reloads, standardized environment variable naming for terminal-aware agents, Auto model selection, and the formal deprecation of the standalone Copilot CLI Agent Host.

If 1.123 was about introducing the agent-first paradigm, and 1.124’s first wave was about multi-session ergonomics, this update is about making agents feel like a first-class, persistent workspace — not a disposable chat window.

Session State Persistence — Your Agents Window Survives Restarts

The most impactful change for daily drivers: VS Code now preserves the complete state of the Agents window across reloads and restarts (#320314). This goes far beyond simply remembering which sessions were open. The full set of restored state includes:

  • Grid layout — session arrangement and sizing
  • Active session — which session was currently selected
  • Pinned sessions — which views were pinned for persistence
  • Per-session layout — terminal, files, and changes views for each session
  • Auxiliary bar visibility — whether the secondary sidebar was open or closed

What this means: A developer working on a complex multi-session investigation — say, refactoring an API while researching a library migration and reviewing PR feedback — can close VS Code at the end of the day and pick up exactly where they left off the next morning. No re-sorting sessions, no hunting for the right context. The “setup tax” that previously accompanied resuming agent work is gone.

Supporting Navigation Polish

Several companion features make the multi-session experience more fluid:

  • Previous/next session keyboard shortcuts (#320331) — navigate through the sessions list without touching the mouse
  • Unified recency history (#320315) — the sessions picker and navigation now share a single recency-aware view, making it easier to find recently used sessions across projects
  • Animated progress indicators (#320318) — session headers now show animated states for in-progress and needs-input, matching the sessions list indicators. When running multiple agents in grid mode, a quick visual scan tells you which sessions are actively processing, which need feedback, and which are idle
  • Chevron toggle for secondary sidebar (#320319) — a new toggle button in the editor title bar (visible in grid mode only) lets you collapse or restore the secondary sidebar, reclaiming screen real estate for agent sessions

AI_AGENT — A Standardized Environment Variable for Agent-Aware Terminals

A small change with big implications for tooling integration: the agent-aware terminal environment variable has been standardized to AI_AGENT (#320127), with the value github_copilot_vscode_agent.

Why this matters: Previously, agent awareness in the terminal was communicated through a VS Code-specific mechanism. By standardizing to a cross-integrator environment variable, Microsoft is signaling that agent-aware terminals are becoming an industry pattern. Tools and scripts can now check for $AI_AGENT to determine whether they’re running in an agent-initiated context, enabling them to adjust behavior — for example, suppressing interactive prompts or piping output differently.

Enterprise takeaway: If your organization builds custom CLI tools, scripts, or development workflows, now is the time to add AI_AGENT detection. As agent-initiated terminal commands become more common across different IDEs and integrators, this standardized variable gives you a single, future-proof check point.

Terminal Steering Uses Session Model

Additionally, terminal steering messages — the prompts agents send to guide terminal operations — now use the session’s configured model instead of the default model (#317059). This is a quality-of-life improvement for teams using different models for different tasks: if you’ve configured a session to use Claude Opus 4.7 for complex reasoning work, the terminal steering will use the same model, maintaining consistency in reasoning quality.

The End of Standalone CLI Agent Host

The Copilot CLI Agent Host can no longer run in empty VS Code windows or with the worktree option when a workspace is open (#318811). Both scenarios now redirect to the Agents window.

This is a strategic simplification. The CLI Agent Host was introduced as a lightweight alternative to the full Agents window, but maintaining two separate agent surfaces created confusion about where to work. By consolidating into the dedicated Agents window, Microsoft creates a single, unified agent experience — with all the multi-session, layout-preserving, keyboard-navigable capabilities described above.

What this doesn’t mean: The Copilot CLI itself is not going away. CLI-based agent interactions remain a core workflow — they’re just landing inside the Agents window where they benefit from the richer UI and session management.

Auto Model Selection and Single-File Diff

“Auto” Model in the Model Picker

The Agents window model picker now supports an “Auto” model option (#311443). When selected, VS Code automatically chooses the best available model for each task based on complexity, context window requirements, and cost considerations. This is particularly valuable for:

  • Teams with mixed model subscriptions — Auto model routes simple tasks to more cost-effective models and complex tasks to premium models
  • Developers who don’t want to micromanage model selection — let VS Code handle the routing
  • AI credit budgets — Auto model can help manage usage-based billing by avoiding expensive models for trivial operations

Single-File Diff Preview

A new Preview setting — sessions.changes.openSingleFileDiff (#320320) — changes the Changes tab to display single-file diffs when selecting a file, rather than showing all changes across the entire workspace at once. For developers reviewing agent-generated code changes — where agents often touch multiple files across a project — this creates a more focused, less overwhelming review experience. Instead of an avalanche of changes, you see one file at a time, in context.

Even More Browser & UX Improvements

  • Single tool call for browser text input (#320182) — agents can now enter text into browser fields in a single tool call, reducing the steps needed for web-based tasks. Agents interacting with web UIs — filling forms, searching documentation — are noticeably faster
  • Improved session list visual hierarchy (#320002, #319372) — better information density and visual cues for scan-reading multiple concurrent sessions

What Has NOT Changed

  • Copilot pricing, licensing, and AI credit model remain unchanged
  • No new model integrations — Claude, GPT, and Gemini support continues as before
  • Extension API is fully backward compatible — existing extensions work without modification
  • Session Sync and Chronicle from 1.123 continue operating as documented
  • All security features (delayed extension auto-updates, sandbox network retry) remain in place
  • WSL agent support from the June 5 release is unchanged

What Organizations Should Do

  1. Update your VSCode deployment. These fixes are in the current stable channel — ensure your team is on 1.124 to benefit from session persistence and the consolidated Agents window experience. For Intune-managed Windows 365 Cloud PCs, update your deployment ring.

  2. Audit CLI tools for AI_AGENT awareness. If your team maintains internal CLI tools, scripts, or dev workflows, add $AI_AGENT detection now. This is the standardized variable that all Microsoft agent tooling will use going forward — getting ahead of it reduces friction later.

  3. Test Auto model with your AI credit budgets. The “Auto” model option can help optimize AI credit consumption. Run a pilot to understand how VS Code routes tasks in your environment and whether it aligns with your cost governance policies.

  4. Enable single-file diff for agent reviews. For teams reviewing agent-generated PRs, enable sessions.changes.openSingleFileDiff (Preview). It makes the review flow significantly more manageable for multi-file agent changes.

  5. Update agent workflow documentation. The CLI Agent Host redirect to the Agents window means any internal documentation or training materials referencing the standalone host should be updated.

The Bigger Picture

This update tells a clear story about where Microsoft is taking VSCode. The session state persistence, AI_AGENT standardization, CLI Agent Host consolidation, and Auto model support are all pieces of a larger puzzle: making agent-assisted development feel as reliable and structured as traditional development.

The key insight from these June 7-8 updates is that Microsoft is investing heavily in the infrastructure of agent work — not just the capabilities. Session persistence turns agent sessions from ephemeral interactions into persistent workspaces. The AI_AGENT variable creates an ecosystem standard for terminal-aware agents. The CLI Host consolidation simplifies the surface area. Auto model selection handles the complexity of model routing at scale.

These are the moves of a platform company thinking about how AI development tools will work in enterprise environments — where reliability, persistence, standardization, and cost management matter as much as raw capability.

At Microsoft Build 2026, the company previewed the Agent 365 SDK and its multi-model agentic security scanning harness (MDASH) — both of which signal that the agent-first development paradigm is not just about writing code faster, but about building a complete platform for secure, governed, enterprise-scale AI-assisted development. Every VSCode update that adds reliability and standardization is a step toward that platform vision.

Ready to optimize your developer tooling at scale? Big Hat Group helps organizations standardize and deploy VSCode with Intune, Windows 365, and GitHub Copilot — from pilot to enterprise-wide rollout. Contact us to discuss your developer productivity strategy.

Big Hat Group is a Microsoft partner specializing in modern endpoint management, Windows 365, and developer productivity tooling.