GitHub pushes autonomous coding workflows deeper into VS Code with March Copilot releases

GitHub’s latest Copilot changelog for Visual Studio Code is less about one isolated feature and more about a clear product direction. In the April 8, 2026 update covering releases v1.111 through v1.115, GitHub shows that Copilot inside VS Code is moving from assisted coding toward a fuller agent workspace, where developers control autonomy levels, attach richer context, and orchestrate multi-step work inside the editor.

The most important addition is Autopilot, now in public preview. GitHub describes it as a mode where agents approve their own actions, automatically retry on errors, and continue until the task is complete without manual approvals. Together with configurable permission levels per session, this gives teams a more explicit way to choose how much control stays with the human and how much shifts to the agent. That matters because agent adoption in software teams often stalls not on raw model quality, but on trust, governance, and interruption cost.

GitHub is also expanding what those agents can do. The update adds integrated browser debugging inside VS Code, allowing developers to set breakpoints, inspect variables, and step through browser sessions without leaving the editor. It also adds nested subagents, so one agent can call another for decomposition of larger tasks. Multimodal interaction is growing too: users can attach screenshots and videos to chat, and agents can return images or recordings that reviewers can inspect in a carousel.

Another important theme is convergence across GitHub’s growing agent surface area. MCP servers configured in VS Code now work in Copilot CLI and Claude agent sessions, which reduces the setup gap between editor-centric and terminal-centric workflows. GitHub also introduced a unified chat customizations editor for instructions, custom agents, skills, and plugins, plus marketplace browsing from the same interface. This suggests the company wants configuration, tooling, and distribution to look more like a platform layer than a collection of separate features.

There are also operational improvements that matter to enterprise teams. The #codebase tool now relies on a single auto-managed semantic index, local MCP servers can run in a restricted sandbox on macOS and Linux, and the /troubleshoot command can analyze agent debug logs directly in chat. Monorepo customization discovery and agent-scoped hooks add more control for larger repos with shared conventions.

Viewed together, these changes make VS Code a more opinionated control plane for coding agents rather than just a place where model output appears. The release does not settle the open questions around cost, review discipline, or reproducibility, but it does show where GitHub is steering the product: toward autonomous, multimodal, tool-rich development loops that blend local editing, CLI execution, MCP connectivity, and structured team customization inside one environment.