The Visual Studio Code 1.127 release is officially out, and it brings a massive shift in how developers and teams will interact with AI agents. This update isn’t just about small quality-of-life improvements; it’s about making AI agents safer, more observable, and significantly better integrated into real-world team workflows, particularly for enterprise environments with strict security and compliance requirements.
Here is a technical breakdown of the most critical new features and why they matter for software engineering teams.
1. Agents Window Preview & Multi-Chat Sessions
What it is
VS Code 1.127 introduces a dedicated Agents window for managing agent activity across your projects. Furthermore, multi-chat sessions allow you to maintain several concurrent conversations and workflows at once.
Why it matters
Historically, interacting with AI in your editor has been highly file-centric. The new Agents window acts as a true control plane for task-centric development (e.g., “refactor this legacy module” or “triage incoming bugs”). Multi-chat sessions allow you to work exactly how developers actually work: handling a production hotfix in one thread while exploring a feature experiment in another.
For teams, persistent agent sessions become shared artifacts. You can now pass an agent session along with a Pull Request so the reviewer understands the AI’s reasoning and constraints.
2. Subagent Credits Visibility
What it is
You now have direct visibility into subagent credit usage, allowing you to see how many tokens or AI credits specific subagents are consuming behind the scenes.
Why it matters
Modern agent architectures rely heavily on subagents specialized for tasks like planning, writing code, or generating tests. Until now, these costs were largely opaque. This new visibility allows teams to monitor costs effectively, compare different orchestration strategies (e.g., planner-heavy vs. direct-execution), and enforce enterprise token budgets. It’s a critical step toward true cost observability for AI in the enterprise.
3. Chat Terminal Command Sandboxing (macOS/Linux)
What it is
When AI agents run terminal commands via chat, those commands are now executed in a sandboxed environment on macOS and Linux, restricting filesystem and network access.
Why it matters
Allowing an AI to run shell commands is incredibly powerful but inherently risky. Sandboxing limits the blast radius of a misconfigured agent or prompt injection attack. This security boundary encourages more “Autopilot” workflows, where developers can allow agents to run setup, build, and test steps autonomously without needing to manually approve every single command, significantly boosting productivity while maintaining least-privilege principles.
4. Deprecation of the Built-In Ollama Provider
What it is
VS Code is officially deprecating the built-in Ollama provider. Users should transition to the official Ollama VS Code extension or other Bring Your Own Key (BYOK) setups.
Why it matters
This move clarifies VS Code’s role as a neutral tooling platform rather than a model vendor. By relying on an extension ecosystem, VS Code encourages standardized BYOK patterns. This is particularly advantageous for enterprise environments that require consistent auditing, logging, and controls across all models they deploy.
5. Integrated Browser Per-Site Permissions & Browser Tools GA
What it is
The VS Code integrated browser now supports per-site permissions (camera, location, sensors). Additionally, Browser Tools for agents have reached General Availability (GA), allowing agents to programmatically interact with the integrated browser.
Why it matters
This effectively turns the editor into a full-stack, UI-inclusive automation surface. With GA Browser Tools, agents can now build a web app, run it, open the browser, click through the UI, and validate functionality. You can now establish closed-loop, end-to-end testing (code → build → run in browser → validate UI) entirely orchestrated by AI inside VS Code.
6. File-Based Delivery for Managed Copilot Settings
What it is
Administrators can now distribute enterprise-managed GitHub Copilot settings via a managed-settings.json file, complementing existing MDM approaches.
Why it matters
For organizations with thousands of machines, ensuring consistent AI configuration is paramount. File-based delivery works seamlessly with configuration management tools (like Ansible or Chef) and provides a way to enforce Copilot policies on machines that might not be enrolled in standard MDM, such as air-gapped systems or Linux dev boxes.
The Bottom Line
With the 1.127 release, VS Code is signaling that AI agents are no longer just “autocomplete on steroids.” They are a manageable, controllable, and secure workforce. Between the Agents control plane, observable costs, sandboxed execution, and robust enterprise configuration options, teams can safely delegate higher-level, autonomous tasks to AI while retaining complete control over their security and budget.
Happy Coding!