GitHub Copilot | VS Code Adds Visual Feedback for Web Agents

GitHub Copilot's latest weekly release brings a particularly useful update for frontend development: Visual Studio Code can now send element-level feedback from its integrated browser directly to an AI agent. Published on August 7, 2026, the roundup also introduces parallel conversations, experimental worktrees in Copilot CLI, improved session management, and new ways to inspect how Copilot handles ongoing development tasks.


GitHub Copilot visual feedback and agent workflows in Visual Studio Code

{getToc} $title={Table of Contents}

Copilot can turn browser elements into precise agent feedback


VS Code 1.132 expands the integrated browser with element-level feedback. Developers can select specific elements on a rendered web page, attach individual comments to them, and send that visual context to the Copilot agent. Multiple elements can be selected and annotated before the message is submitted.


For frontend workflows, this creates a more direct connection between what a developer sees in the browser and what the agent needs to change in the code. Instead of describing a button, navigation item, card, or layout section only through text, the developer can point to the exact interface element and explain what needs attention.



Side questions can stay separate from the main agent task


VS Code also adds the /btw command for asking a side question without interrupting the agent's current turn. The side conversation shares context and prompt cache with the primary chat, allowing developers to ask about work already in progress while keeping the main task intact.


Text from an existing response can also be selected and used as the context for a follow-up question. This can be useful during larger frontend changes where a developer may want clarification about a generated implementation, dependency, or design decision without redirecting the agent away from the original task.


Copilot CLI adds isolated worktrees and concurrent sessions


Copilot CLI is receiving more tools for parallel development. A new Sessions sidebar makes it easier to keep several active conversations organized and move between them, while the experimental /worktree command creates an isolated Git worktree and starts a separate conversation inside it.


This gives developers another way to test an alternative implementation without disturbing the current workspace. The CLI also adds /rewind support without Git, allowing Copilot-modified files and conversation state to be restored while later user edits are preserved, and its timeline can now display live tool-call durations.


The Copilot app improves parallel work and session visibility


The GitHub Copilot app receives its own session improvements. Auto model selection now shows which model completed a request and can expose AI credit and cache information when that data is available. Shared sessions can be opened directly, while /side provides another way to explore a parallel question without disrupting the main task.


GitHub has also updated the Impeccable skill with support for mobile interfaces, setup diagnostics, and more focused design reviews. Combined with faster session startup and switching, the weekly release continues moving Copilot toward workflows where several related development and design tasks can remain active without being forced into a single conversation.


IMPORTANT: The /worktree command in GitHub Copilot CLI is currently experimental. Treat isolated agent workflows as development tools and review generated changes before merging them into a production branch.{alertWarning}

Daisuki's Take: What This Means for Web Designers


Element-level browser feedback is the most interesting part of this release for web designers because it reduces one of the recurring problems with prompt-based frontend editing: explaining exactly which part of an interface needs to change. Visual selection gives the agent a much more precise target.


That could make iterative UI work more natural when adjusting spacing, components, responsive layouts, or visual details. Side conversations and isolated worktrees also provide useful ways to explore alternatives without losing the context of the primary implementation.


We would still treat the browser annotation as guidance rather than proof that the final implementation is correct. Responsive behavior, accessibility, generated code, and cross-browser results should still be reviewed directly, but the path from spotting a visual problem to asking Copilot to address it is becoming noticeably shorter.



Sources and Recommended Links