VS Code 1.145 mixes Claude and Copilot in one session, and previews remote agent dev servers in the window

VS Code shipped 1.145 on November 4. The version number reached three digits, and the contents finally stopped being “a few more completions”: the theme this time is multiple AI agents collaborating inside one session.

Nested sessions can switch harness

Previously a nested session inherited its parent’s harness, the agent engine behind it. Getting a second opinion from another agent meant starting a separate session, which lost the grouping.

In 1.145, nested sessions choose their own harness:

• Ask Claude to review a plan Copilot wrote
• Give Codex the same task and compare both side by side
• Each chat keeps its own model, agent and permission level; changing one does not affect the others
• Moving a chat keeps its history; forking keeps its configuration

The significance is that multi-model moves from “a different dropdown” to “a conversation in the same context”. Comparing two models used to mean duplicating the request and maintaining two contexts. Now they share one tree.

Remote agents: the preview window connects

If an agent session runs on a remote host or in a container, the integrated browser now loads pages through that host.

So when the agent starts a dev server remotely, you open it in the Agents window, validate the change and give feedback, exactly as in a local session. Turn it on with workbench.browser.enableRemoteProxy. Canvases work too, with chat.canvases.enabled.

For me this matters more than harness mixing. I work on Windows while the projects actually run elsewhere. “The agent changed something on the remote box and I cannot see it” used to mean guessing from screenshots and logs.

The rest

• Cron for automations: when the presets are not enough, schedule recurring agent tasks with expressions
• Agent verbosity: keep chats to conclusions, or expand reasoning and tool calls
• File exclusions that keep files searchable: fold noisy folders away without losing the ability to find things
• Several MCP and extension fixes
• A surprise: a platform-jumping game starring Blobby, the VS Code pet

My take

I do not buy the “AI replaces the editor” line, but I do believe the editor is becoming an agent control room. Every change in 1.145 points the same way: selectable harnesses, remote previews, scheduled tasks, adjustable verbosity.

None of it is “smarter”. All of it is “more controllable”. For people who write code in it all day, controllable beats smart, because you do not need it to write everything for you. You need to be able to catch it when it drifts.

And that little game. A release that puts agent orchestration at the core, with an official easter egg that tells you to take a break, is the most honest release note of the year.