VS Code 1.135: Rubber Duck review, persistent agents, and token clarity
Microsoft shipped VS Code 1.135 with four changes that push the editor further toward a full agentic development environment. If you use GitHub Copilot agents or Claude agents inside VS Code, each of these touches your daily workflow.
Rubber Duck: a second opinion on agent work
The headline feature is Rubber Duck, an AI review layer that applies a second model to check the output of the first. The logic behind it: a single AI agent working alone can develop blind spots that are hard to catch from inside its own reasoning. Rubber Duck externalizes that check.
It borrows the name from the classic developer technique of explaining your code to an inanimate object to catch your own mistakes. Here, the rubber duck talks back.
⚙️ Persistent agent sessions
Agent sessions now run in their own process and persist across windows and applications. That is a meaningful architectural shift. Previously, closing a window could orphan an in-progress agent task. Now the session survives the surface.
The practical effect: the editor becomes the control plane where agent work is directed and reviewed, rather than a single-window execution context you have to keep open.
️ Agent Host Protocol (AHP)
VS Code 1.135 introduces Agent Host infrastructure built around the Agent Host Protocol. The release positions AHP as the underlying layer that lets agents run reliably outside the immediate editor context. Details on the full API surface were not published in the release notes at the time of reporting.
Clearer token usage
Token visibility gets an upgrade. The release surfaces clearer information about token consumption during agent sessions, which matters for anyone watching API costs on Copilot or third-party model integrations. Knowing what a session costs before it finishes is a basic operator need that has been missing from most AI coding tools.
The operator read
Rubber Duck and persistent sessions are the two features worth testing first. The second-opinion review addresses a real failure mode: AI agents that confidently produce wrong output because nothing in their loop pushes back. If you run long agentic tasks in VS Code, both features reduce the cost of catching errors late.
VS Code 1.135 is available now.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
