VS Code 1.135 Gives AI Coding Agents a Second Opinion

Microsoft's latest update to Visual Studio Code, version 1.135, introduces an experimental feature called Rubber Duck to provide AI coding agents with a second opinion. The feature, inspired by the traditional programming technique of explaining code to a rubber duck, utilizes a different AI model to review the primary agent's plans, code, and tests.

This aims to catch edge cases and gaps that the original model may have missed, ultimately surfacing potential production bugs before they occur. Microsoft describes Rubber Duck as a complementary model that offers a "second opinion from a complementary model on the agent's work," targeting the challenge of AI models having blind spots in their reasoning.

In addition to Rubber Duck, VS Code 1.135 also introduces an Agent Host built on the Agent Host Protocol, allowing developers to connect to the same agent session from multiple VS Code windows. This enhances session continuity, especially for those who work across multiple devices or need to check in on long-running agent tasks without losing context.

The Agents window has also been redesigned with a single-pane layout, interactive control pills, and a more responsive diff visualization. For developers concerned about AI spend, the update includes a token breakdown display for each chat response, providing clearer insights into model usage. While this release addresses verification pressure with additional generation, it's essential to remember that Rubber Duck remains an experimental feature and a second model reviewing the first one's work doesn't guarantee correctness.

The development of agentic coding tools is evolving, with this update focusing on building structure around the speed of AI coding assistants. The direction suggests a future where AI agents are directed and reviewed through a durable and portable session infrastructure, ultimately enhancing the overall coding experience. VS Code 1.135 is now available through the standard update channel.