GitHub published "GitHub Copilot in Visual Studio Code, June 2026 releases" on July 8, 2026, covering VS Code v1.123 through v1.127. The important point is not one new button. It is that VS Code is becoming a fuller agentic development workspace.
The first major update is the integrated browser. GitHub says agentic browser tools are now generally available and enabled by default. Agents can navigate pages, inspect content, capture screenshots, and validate web apps directly inside VS Code. That is especially important for front-end and full-stack workflows because an agent is no longer limited to editing code. It can see the result, gather page context, and use screenshots as evidence.
The second update is parallel sessions and chats. The Agents window can keep separate agent tasks visible and active at the same time, and a single session can contain multiple chats. In practice, that allows a larger engineering task to be split into implementation, review, testing, and documentation workstreams instead of compressing everything into one chat thread. For long tasks, that model looks much closer to real engineering management.
The third update is cost visibility. GitHub says users can see total session cost and inspect credit usage for individual subagent sections when work is delegated. This is a necessary step for mature agent workflows. Once AI can split work, call subagents, and use longer context windows, teams need to know not only whether the answer was good, but how much each delegated step cost.
The fourth direction is model providers from the Marketplace. VS Code's Language Models editor can help developers discover model provider extensions, install them from Marketplace, and tune context size and reasoning effort from a unified picker. Coding agent tools are moving from one fixed model toward task-based model and reasoning choices. Different work requires different tradeoffs across speed, context, cost, and reliability.
Autopilot also improved. GitHub describes it as a permission level that lets agents act without checking in at every step, with better judgment about whether a task is actually finished. That is powerful but sensitive. The more independently an agent can move, the more important permissions, cost views, session history, gutter feedback, and pull request generation become.
Taken together, these VS Code updates combine browser validation, parallel agent work, cost governance, model choice, less interrupted Autopilot behavior, and managed Copilot settings. Agentic coding is no longer just chat-based code completion. It is becoming an observable, governable, multi-workstream development environment.
