Visual Studio Code 1.131 landed on July 29, 2026, and it’s a release that sharpens the edges of the AI-assisted development experience. Where recent releases built the agent host infrastructure, 1.131 focuses on making agents more transparent to work with, adding voice input natively, and giving Markdown a first-class hybrid editing surface inside the Agents window.

Let’s break down what shipped.

Subagent Visibility: See What Your Agents Are Doing

The headline feature for anyone running complex agent workflows. When an agent delegates tasks to subagents — parallel context windows that handle parts of a larger task — you previously had to open each subagent’s conversation to check its progress. Not anymore.

In the Agents window, the main conversation now shows for each running subagent:

  • The model the subagent is using
  • How long it’s been running
  • The tool it’s actively calling right now

Click a running subagent to open its conversation in a separate chat for a detailed review, while the parent conversation stays available. This is the kind of feature that sounds small until you’ve kicked off a multi-step refactoring with three subagents running in parallel and need to know which one is stuck on a terminal command and which one is about to finish.

Combined with the subagent credit visibility introduced in 1.127, you can now audit both what your subagents are doing and what they cost — without leaving the main conversation.

Built-in Dictation Across VS Code (Experimental)

Settings: dictation.enabled, dictation.showTranscript, dictation.experimental.llmCleanup

You no longer need the VS Code Speech extension to dictate text. Built-in dictation works across three surfaces — chat inputs, text editors, and the integrated terminal — with a single shared speech session and microphone selection. This prevents the classic problem of accidentally dictating into two surfaces at once.

Key details:

  • Offline Nemotron model: Audio is processed on your device. The model downloads on first use and stays local. Nothing leaves your machine.
  • LLM cleanup: With dictation.experimental.llmCleanup enabled, Copilot refines your transcript in real time — adding formatting and stripping filler words. If the cleanup service is unavailable, VS Code keeps the raw transcript.
  • Live transcript toggle: dictation.showTranscript controls whether the transcript appears as you speak.

Supported platforms: Windows x64/Arm64, macOS on Apple silicon, Linux x64/Arm64 (glibc 2.34+), and remote workspaces (transcription runs locally on the client).

Not supported: VS Code for the Web, Intel-based Macs, 32-bit and Arm32 systems.

For developers with repetitive strain injuries or anyone who prefers voice input for long-form documentation, this removes the need for OS-level dictation tools that aren’t tuned for coding workflows. It’s experimental, but it’s a meaningful step toward multimodal input in the editor.

Hybrid Markdown Editor (Experimental)

Setting: workbench.editor.markdownDefaultEditorInAgentsWindow

A new Markdown editor designed specifically for the Agents window. It lets you:

  • View Markdown files with rendered formatting
  • Edit them in-place
  • Add comments that an agent can act on

Use Reopen Editor With to switch between the traditional text editor and this new hybrid editor — both in the Agents window and in regular editor windows.

This is built for the workflow where an agent generates or modifies Markdown documentation and you want to review, tweak, and annotate it without leaving the agent’s context. Instead of jumping between a preview pane and a raw editor, you get a single surface that handles both.

VS Code Pet (Experimental)

Enter /vscode-pet in chat to meet a new companion. It’s highly experimental and delightfully unnecessary — exactly the kind of easter egg that makes release days fun.

Agent Host: Steady Progress

The agent host — the dedicated process running agent sessions (Copilot, Claude, Codex) based on the Agent Host Protocol (AHP) — continues its progressive rollout. Enable chat.agentHost.enabled and select a harness from the dropdown to opt in.

The Copilot agent is powered by the Copilot SDK, which means its behavior aligns with the Copilot CLI, the standalone GitHub Copilot app, and other Copilot products. Sessions run in their own process and can be connected to from multiple VS Code windows simultaneously.

Accessibility: Better Terminal Screen Reader Support

Setting: terminal.integrated.accessibleViewPreserveCursorPosition

Set to always to preserve cursor position in the terminal Accessible View even when new output arrives. Previously, new terminal output could disrupt screen reader navigation. Now, terminal live updates use a non-interrupting ARIA status announcement instead of an assertive alert, so screen readers can keep reading at the user’s pace without repeated interruptions.

Terminal: Resize Overlay Control

Setting: terminal.integrated.resizeDimensionsOverlay.enabled

If the columns-by-rows overlay that appears when resizing the terminal is distracting, you can now disable it. Changes apply immediately to open terminals — no restart needed.

Python Environments: Default Experience GA

Python Environments is now the default environment-management experience in both VS Code Stable and Insiders, with the rollout reaching 100% of users. This release also brings performance improvements:

  • Conda discovery deferred until needed, reducing startup overhead
  • Concurrent environment scans consolidated to avoid redundant work
  • Pylance uses last-known interpreter during full environment refreshes, so you keep IntelliSense while environments update

For Python developers, this means faster project startup and fewer “waiting for environment detection” stalls.

Community Contributions

This release saw notable community work, particularly from Simon Siefke who fixed six separate memory leaks across the task service, runtime extensions editor, terminal process manager, debug model, terminal service, and main thread terminal service. Other contributors:

  • Arthur Cnops (@accnops): Voice passive push-to-talk hands-free narration fix
  • Brandon Waterloo (@bwateratmsft): When clause context key for detecting if an extension is installed and enabled
  • Kaidesuyo (@Kaidesuyoo): Performance regression fix for modern UI styles
  • Mir (@mirimadahmed): Voice barge-in regression fix
  • Piyush Madan (@piyushmadan): Resolve execution subagent model by exact CAPI ID before family fallback
  • Jori Huisman (@SixFive7): Honor Explorer’s minItems for Windows taskbar jump list
  • soreavis: Git — resolve active notebook for Open Changes/Open File

Issue tracking contributors: John Murray (@gjsjohnmurray), RedCMD (@RedCMD), Andrii Dieiev (@IllusionMH), and Alberto Santini (@albertosantini).

What’s Not in This Release

No deprecated features or settings — a clean release on that front.

The Bigger Picture

VS Code 1.131 is a polish release with three distinct themes:

  1. Agent transparency: Subagent visibility means you can watch what your parallel agents are doing in real time — model, duration, active tool — without context-switching.
  2. Multimodal input: Built-in dictation with an offline model is a meaningful step toward voice-native development workflows, especially for accessibility.
  3. Documentation as a first-class agent surface: The hybrid Markdown editor recognizes that agents increasingly generate and modify documentation, and gives that workflow a dedicated editing experience.

If you’ve been enabling the agent host and running multi-agent workflows, 1.131 makes that experience substantially more observable. If you haven’t tried dictation yet, enable dictation.enabled and give it a spin — the offline model means there’s no privacy tradeoff to experiment with.

VS Code 1.131 is rolling out gradually. Use Check for Updates in VS Code to get it immediately, or download the nightly Insiders build for early access to upcoming features.