VS Code 1.134 AI Agent Sessions: What’s New

Yes, Visual Studio Code 1.134 puts AI agent sessions front and center, making them a core part of the interface rather than a hidden experiment. The update adds a dedicated Agent Sessions view, a new agent mode for Copilot, and a streamlined way to manage multiple concurrent AI tasks directly from the primary sidebar. This article explains what changed, how the new session workflow works, and what it means for your daily coding.

Background: From Chat to Agent Sessions

VS Code has had a chat panel for years, but the shift to agentic AI changed the game. Instead of one-off Q&A, Copilot can now plan, edit multiple files, run terminal commands, and verify results. Earlier versions treated each chat as an isolated thread; 1.134 introduces a formal session layer that tracks these multi-step agent runs as first-class objects.

The key difference: a session is not just a conversation history. It includes the agent’s plan, the files it touched, the terminal commands it ran, and the final diff. You can switch between sessions, pause one, resume another, and compare what each agent did—all without losing context.

How Agent Sessions Work in 1.134

When you start a new agent task (for example, by pressing Ctrl+Shift+I and choosing “Agent”), VS Code creates a session. The session appears in the new Agent Sessions view, which sits in the same sidebar as your file explorer. Each session shows:

• The user prompt that started it
• The agent’s plan (steps it intends to take)
• A list of files changed or read
• Terminal commands executed
• A status indicator (running, waiting for input, completed, or failed)

You can rename a session, delete it, or click it to reopen the full chat transcript. The view also shows a summary of the final diff, so you can quickly review what the agent did without opening every file.

Agent Mode vs. Chat Mode: What’s the Difference?

VS Code now clearly separates two modes:

1. Chat mode – answers questions, explains code, suggests snippets. It does not edit files unless you explicitly ask and approve each change.
2. Agent mode – works autonomously on a task: it can edit multiple files, run commands, and iterate until the job is done. It asks for approval before executing terminal commands or making large changes.

The 1.134 update makes this distinction more visible. The mode selector is now a prominent toggle at the top of the Copilot panel, and each mode has its own icon. This reduces the confusion that arose when the agent was buried inside the chat interface.

Managing Multiple Sessions: The New View

Before 1.134, running several agent tasks meant juggling multiple chat tabs. Now, the Agent Sessions view gives you an overview. You can:

• Start a new session with a single click
• Switch between active sessions without losing state
• See which sessions are still running (useful when you have a long build or test loop)
• Clear old sessions to keep the list tidy

This is especially helpful for large refactors where you might have one agent migrating a module while another updates tests. You can monitor both simultaneously.

Common Problems and Workarounds

Users have reported a few issues with the new session system:

• Session list doesn’t refresh – sometimes the view needs a manual refresh (click the refresh icon in the view header) or a window reload.
• Agent gets stuck waiting for approval – if a terminal command requires confirmation and you don’t notice, the session appears hung. Check the status icon; it will show “waiting for input.”
• Old sessions consume memory – each session holds the chat history and file diffs. If you have many large sessions, VS Code may slow down. Delete old ones or use the “Clear All” action.
• Session not saved after restart – by default, sessions are stored in memory. To persist them, enable the setting chat.agentSessions.persist (this is a new setting in 1.134).

How to Choose Between Chat and Agent

Use chat mode for quick questions: “What does this function do?” or “Explain this regex.” Use agent mode for multi-step tasks: “Refactor this module to use async/await” or “Add error handling to all API calls.” The agent can handle the repetitive parts, but you should always review the final diff.

A good rule of thumb: if the task would take you more than five minutes of manual editing, consider the agent. For anything that touches critical code (auth, payments, data migration), review every change carefully.

Safety and Best Practices

Agent sessions can make sweeping changes. To stay safe:

• Always work in a Git branch before starting an agent task. That way you can revert easily.
• Review the plan before letting the agent execute. The plan appears at the start of the session.
• Approve terminal commands individually. Don’t enable “auto-approve” unless you trust the task completely.
• After the agent finishes, run your tests and check the diff in the session view.

VS Code also adds a visual indicator when an agent is about to modify a file outside your workspace—it will ask for explicit permission.

Frequently Asked Questions

Is the Agent Sessions view available in all versions of VS Code?

No. It’s a feature of the 1.134 release, which is currently in the Stable channel. If you’re on an older version, update via Help > Check for Updates. The feature also requires the latest Copilot extension.

Can I use agent sessions without GitHub Copilot?

No. Agent sessions are a Copilot feature. The view will appear only if you have the Copilot extension installed and are signed in. If you don’t have Copilot, you won’t see the Agent Sessions view.

How do I share an agent session with a colleague?

You can export the session transcript as a JSON file (right-click the session and choose “Export”). Your colleague can import it via the same menu. However, the imported session is read-only—they can’t resume it, only view the history.

Does the agent session count against my Copilot usage limits?

Yes. Each agent run consumes premium requests, just like regular Copilot chat. The exact limits depend on your plan. You can check your usage in the Copilot status bar icon.

Why is my agent session “stuck” on a step?

It’s likely waiting for your approval. Open the session and look for a prompt asking to run a terminal command or edit a file. You can also click “Cancel” to stop the agent and keep the partial changes.

Can I run multiple agent sessions at the same time?

Yes. Each session runs independently. However, be aware that concurrent agents may conflict if they edit the same files. It’s safer to run them on different parts of the codebase.

Alternatives and What’s Next

If you don’t use Copilot, other extensions like Continue or Cline offer similar agentic features, but they don’t integrate with the native Agent Sessions view. The new session system is a major step toward making AI assistance a structured part of the editor, rather than an add-on. Future updates are likely to add session sharing, better conflict resolution, and deeper integration with the debugger.

For now, the Agent Sessions view is a solid addition that makes multi-step AI work more manageable. Try it on a small refactor first, and you’ll quickly see the benefit of having a clear record of what the agent did.

Related

• Convert Table to Text in Word: Step-by-Step Guide
• How to Remove Extra Space in Word: Step-by-Step
• How to Print Background Color in Word 2013