AI coding assistants like GitHub Copilot are powerful partners, capable of refactoring entire codebases or scaffolding new applications from a single prompt. But with great power comes the potential for chaos. A slightly ambiguous instruction or a misguided prompt can send an AI on a multi-file editing spree, leaving a developer with a tangled mess that’s difficult to untangle. Addressing this very problem, Microsoft’s latest update for Visual Studio Code (v1.103) introduces VS Code chat checkpoints, a groundbreaking feature that acts as a safety net for AI-driven development.
This new capability allows developers to instantly roll back both their workspace and their Copilot conversation to an earlier, stable state, effectively providing an “undo” button for complex AI interactions. Announced in early August 2025, the update is part of a broader push to make AI interactions safer and more manageable for developers. It arrives alongside significant upgrades to Copilot’s underlying tooling and the integration of OpenAI’s powerful new GPT-5 model, signaling a major step forward in the evolution of AI-assisted coding and development trends.
For professionals who rely on VS Code, this release is less about introducing flashy new AI abilities and more about building the trust and reliability needed to use existing AI tools for more ambitious tasks. It acknowledges that as AI becomes more autonomous, developers need robust controls to manage its actions and confidently experiment without fear of irreversible mistakes.
A Safety Net for AI-Driven Edits: Chat Checkpoints Explained
The headline feature of the VS Code 1.103 update is Chat Checkpoints, a system that brings snapshot-style versioning to AI chat sessions. Think of it as a session-aware undo button that works across your entire workspace. When you’re deep in a conversation with Copilot and it makes a series of edits across multiple files, a single Ctrl+Z isn’t enough to revert the changes cleanly. Chat Checkpoints solve this by creating automatic snapshots at key moments in your interaction.
According to the official release notes, if you realize a series of AI-driven edits has taken your project in the wrong direction, you can simply find the last “good” point in your chat history and select “Restore Checkpoint.” This action reverts all modified files in your workspace to the state they were in at that moment and simultaneously rolls back the chat history. This makes it easy to backtrack from a failed experiment and try a different prompt.
Should you change your mind again, a “Redo” option is available to re-apply the changes you just undid. This feature is enabled by default and provides a much-needed safety layer, encouraging developers to use Copilot for more ambitious tasks like large-scale refactoring, knowing they have an easy way to reverse course if the AI misinterprets their intent.
Under the Hood: How Checkpoints Integrate with Your Workflow
Unlike traditional version control systems like Git, Chat Checkpoints are not meant for long-term history tracking. Instead, they are designed for immediate, in-session recovery. The feature is lightweight and operates seamlessly in the background without requiring any manual action from the developer.
To enhance transparency, developers can enable the chat.checkpoints.showFileChanges setting. This displays a list of all files that were modified at the end of each chat request, along with the number of lines added or removed. This context makes it much easier to identify the exact point where an AI-driven edit went wrong and decide which checkpoint to restore.
While Git commits are for saving deliberate, verified changes, checkpoints are for managing the fluid, experimental nature of AI-assisted coding. They provide a more granular and immediate rollback option that is perfectly suited for the iterative process of prompting and refining AI-generated code. You can find more details in the official documentation.
Expanding the Toolkit: Major Updates to MCP
The v1.103 release also brings significant improvements to the Model Context Protocol (MCP), the system that allows Copilot’s agent mode to discover and use specialized tools from extensions. These updates are crucial for developers building more complex, multi-tool AI workflows.
A key limitation has been the 128-tool cap on a single chat request. Previously, if a developer had multiple MCP servers with many tools installed, they would hit this limit and have to manually deselect tools to proceed. This update introduces an experimental “tool grouping” mode. When the number of tools exceeds the threshold, VS Code now automatically groups them, and the AI model is given the ability to activate and call an entire group at once.
Other MCP enhancements include:
-
- Revamped Tool Picker: The interface for selecting tools has been completely redesigned with a new “Quick Tree” component, making it easier to manage large sets of tools with expand/collapse controls and sticky scrolling.
- Server Autostart: A new setting,
chat.mcp.autostart, automatically restarts MCP servers after configuration changes, removing the need for manual refreshes. - Trust Prompts: For enhanced security, VS Code now prompts users to confirm they trust an MCP server before it runs for the first time after an update.
These updates to the underlying protocols are essential for building the next generation of AI assistants and are part of the core technologies of 2025 that are making AI a more integrated part of the development environment.
The Power of GPT-5 and Other AI Enhancements
This VS Code update also officially brings support for OpenAI’s latest and most powerful model, GPT-5, along with its smaller variant, GPT-5 mini. Developers with paid GitHub Copilot plans will be able to select GPT-5 in the model picker, unlocking its advanced reasoning and decision-making capabilities for more complex coding tasks.
Alongside the new model, several other AI-centric features have been introduced in preview:
-
- AI Usage Stats: A new feature that records the percentage of code in a file that was inserted by AI versus typed manually by the developer, giving a tangible metric for AI’s contribution.
- Task Lists for Agent Mode: When using Copilot in agent mode for high-level tasks, a new task list appears in the chat panel, showing the agent’s plan and ticking off sub-tasks as they are completed.
- KaTeX Rendering: For developers working with mathematical formulas, chat responses can now render expressions using KaTeX when the feature is enabled.
Beyond AI: Notable General Improvements in v1.103
While the focus of this release is heavily on AI, several other significant improvements have been made that will benefit all VS Code users.
Most notably, VS Code now supports Git worktrees, a feature that has been highly requested by the community since 2019. Git worktrees allow a developer to have multiple branches of the same repository checked out simultaneously in different folders, which is incredibly useful for working on different features or fixing bugs without disrupting the main workflow.
Other key updates include:
-
- TypeScript 5.9 Support: The editor now supports the latest version of TypeScript, which includes a new, expandable UI for showing detailed type information on hover.
- Terminal Improvements: The settings for automatically approving terminal commands have been consolidated and improved with better support for regular expressions. The inline chat also now features more accurate detection of the active shell, leading to more relevant command suggestions.
Practical Impact
For developers, the practical impact of the VS Code 1.103 update is a significant reduction in the risk associated with AI-driven coding. The introduction of Chat Checkpoints creates a much-needed safety net, encouraging developers to experiment more freely with Copilot’s agent mode for large-scale changes. This feature transforms AI from a tool you use with caution to a more robust partner you can trust to attempt complex tasks, knowing you can easily revert if the outcome isn’t right. This shift from capability to reliability is a sign of a maturing AI development ecosystem.
Frequently Asked Questions (FAQ)
1. What are Chat Checkpoints in VS Code?
Chat Checkpoints are a new feature in VS Code v1.103 that automatically saves snapshots of your workspace and chat history. This allows you to restore your project to a previous state if an AI-driven edit from GitHub Copilot produces undesirable results across multiple files.
2. Do I need to enable Chat Checkpoints?
No, the feature is enabled by default in VS Code version 1.103 and later. You can, however, disable it using the chat.checkpoints.enabled setting if you choose.
3. How is a checkpoint different from a Git commit?
A checkpoint is a temporary, in-session snapshot designed for quick rollbacks during an experimental AI chat. A Git commit is a permanent, deliberate save to your project’s version history. Checkpoints are for undoing mistakes, while commits are for saving progress.
4. What is MCP and why are the updates important?
MCP, or Model Context Protocol, is the system that allows Copilot to use specialized tools in agent mode. The latest updates, like tool grouping, allow Copilot to handle a much larger number of tools at once, making it more powerful and versatile for complex, automated tasks.
5. Does this update mean I can use GPT-5 in VS Code now?
Yes, support for GPT-5 and GPT-5 mini is rolling out to all paid GitHub Copilot plans. Users will be able to select it from the model picker in the Copilot Chat view to access its advanced reasoning capabilities.