Visual Studio Code 1.119 finally stops AI agents from blindly reading your browser tabs by forcing explicit sharing controls and adding a network-friendly sandbox mode that kills those endless approval prompts. You can now track exactly where tokens vanish using OpenTelemetry tracing, while an experimental background manager handles progress lists so the main model actually focuses on coding instead of busywork. Markdown preview switching gets dedicated toolbar buttons for instant toggling, and webviews shift to CSS anchor positioning to fix the laggy panel dragging that has annoyed developers for years. A full migration to TypeScript 7 cuts typechecking times down to a fraction of their previous length, though teams should quietly prepare for Edit Mode to get permanently deleted in version one point two five.

Visual Studio Code 1.119 brings smarter agent controls and faster Markdown previews to your workflow

The latest Visual Studio Code 1.119 update shifts focus from raw feature dumping to actually making AI agents behave like reliable coworkers instead of chaotic interns. This release tightens security sandboxes, trims token waste on long tasks, and finally makes switching between Markdown source and preview a click away. Developers who have wrestled with agent prompts or watched their IDE choke on heavy webviews will find several practical fixes here.

Agent controls and sandbox tweaks

Agents that can peek inside a live browser tab sound convenient until they start reading sensitive account pages or triggering unwanted logins. The new sharing model forces developers to explicitly attach tabs through the context picker or drag-and-drop interface, which keeps private data locked away from automated scripts. A prompt now appears when an agent requests access to an open tab, and it even suggests reusing existing tabs on the same domain instead of spawning fresh windows that clutter the workspace. The sandbox settings also get a practical update for teams that need network access without sacrificing file system protection. Setting chat.agent.sandbox.enabled to allowNetwork removes constant approval prompts for package installs or dev server calls while keeping local files safe from rogue tool executions.

Observability and token optimization

Long agent sessions usually end with a mystery about where tokens vanished or why a multi-step task stalled out of nowhere. This update routes Copilot CLI, Claude agent, and local agent activity through OpenTelemetry traces that map every tool call and model interaction into a single connected timeline. Developers can now point the otel.otlpEndpoint setting to any compatible collector like Aspire Dashboard to track latency and cache breakdowns without guessing which step caused a bottleneck. Token waste also takes a hit when managing complex task lists. A new experimental background agent handles todo list updates using a lighter model, leaving the main reasoning engine free to focus on actual code generation instead of constantly rewriting progress trackers. The feature stays disabled by default since manual tool overrides will bypass it anyway.

Markdown preview toggles and webview performance

Switching between raw markdown source and the rendered preview used to hide behind obscure menu commands that most developers never found. The new toolbar buttons make the toggle instantly accessible, and the settings panel now groups all markdown language features under a single Preview subsection for easier discovery. Behind the scenes, the webview engine finally abandons JavaScript-based positioning in favor of CSS anchor placement. This shift eliminates the laggy relayouts that used to happen when dragging panels or resizing windows with multiple active views open. The browser now handles coordinate math directly, which also fixes those persistent bugs where embedded panels would drift out of alignment on web builds.

TypeScript migration and edit mode retirement

Typechecking speed matters when developers need rapid feedback loops during heavy refactoring or agent-assisted debugging sessions. The core codebase and all built-in extensions now run on TypeScript 7, which slashes type checking time from twenty-two seconds down to four for the Copilot extension alone. Faster compilation means agents and human editors alike get quicker error reports without waiting through unnecessary build delays. Meanwhile, the controversial Edit Mode feature finally gets a clear expiration timeline after lingering in preview status since version one point one zero. The setting remains available through version one point two five before being permanently removed, giving teams enough time to adjust their workflows or migrate to standard chat commands.

Downloads

Windows

x64  Arm64 

Mac

Universal  Intel  silicon

Linux

deb  rpm  tarball  Arm  snap

Grab the update when ready and test the sandbox settings against your current agent workflows. The token tracking features alone will save headaches on long debugging runs. Happy coding.