Transcription
The latest VS Code release brings a whole new experience for managing your agents, whether local, cloud, background, or even cloud or codecs. Let's dive in.
The teams landed a lot of new changes to the agent experience in VS Code, so you can use multiple agents to tackle all of your coding tasks. There's a new agent sessions welcome experience that opens when you start up VS Code. It gives an overview of your agent sessions so you can jump right into them or see them at a glance and kick off a new agent session right away. You can also open the agent session sidebar at any time by clicking the chat icon in the command center. To see the agent sessions page at startup, you'll want to set the workbench startup editor setting to agent sessions welcome page as this is still an experimental setting.
When I go to kick off an agent session, I can work with multiple types of agents. I can use a local agent to work interactively, a background agent for changes isolated in a git work tree, a cloud agent for changes running in the cloud that result in a PR, and even other providers like Claude or Codeex as local agents. Let's start with a local plan agent that will research and create a multi-step plan and ask to suggest improvements for making my app more accessible. We'll kick that off while I continue my work with other agents.
I'm going to start a background agent so that it will work asynchronously via the co-pilot CLI in a git work tree. And this is an app I have for meal planning that I want to add a new page for likes and dislikes. Throughout this, I can track and check on my agents via the counter at the top of the editor. Filter by session status and maximize chat input while agents continue working.
Next, I'm going to delegate some work to the cloud agent. And you can now use Copilot, Claude, or Codeex cloud agents, all with your single GitHub C-Pilot subscription. I'll stick with Copilot and ask it to make sure the readme is updated with the latest functionality. This will then kick off the co-pilot coding agent to create a pull request and if I had any pending changes, it would walk me through managing that. Once the pull request is created, I can view it using the GitHub pull request extension, then return to the agent sessions list to track my sessions in progress.
The local plan agent finished its initial plan for improving accessibility. So now I can choose to keep iterating or start implementing the plan. Now I'll jump back to the background agent I kicked off to add a new page. I can then review the changes that were generated on the git work tree and after reviewing choose to apply the changes to my workspace which adds those changes in the source control view. I can then decide to run a co-pilot code review from here, walk through the comments and when ready I can use copilot to generate a commit message for a full copilot flow from prompt to push.
One last feature that I will leave you with is the ability to run parallel sub aents. I'm going to kick off a local agent session asking to use sub aents to analyze my project's error handling across the front end, backend, and database layers. While that runs, a little bit more about sub aents. A sub aent is a context isolated agent that runs independently from your main chat session. This is beneficial because it means it keeps your context cleaner by dispatching jobs out to a sub agent. Then it just brings the result back to the main chat. And now you can run parallel sub aents making for even quicker results. So we can see here that multiple sub aents have spun up each tackling their specific part of the project and are all running at once with self-contained context that we can still follow.
We're so excited to be on this journey of making VS Code your agent UX. The team's constantly shipping new features and would love any feedback. So, make sure to drop your thoughts in the comments or file an issue directly on the VS Code repo. As always, happy coding.