GitHub Copilot’s Latest Update Bets on Model Choice, Not Model Loyalty

GitHub's recent Copilot update, announced on August 13, reflects the growing trend in AI model selection. Instead of sticking with a single model, development teams now choose the most suitable model for each task. The update brings two new models: Kimi K3 and MAI-Code-1.1-Flash. Both models augment existing options, enabling developers to select their preferred model at any point during a task.

Microsoft's Visual Studio Code 1.133 update further enhances model-switching capabilities. Now, developers can switch between Claude BYOK and built-in Copilot models within a single session, without losing context or restarting the session. This flexibility allows developers to allocate tasks to different models seamlessly.

However, this convenience comes with a drawback. Switching models dynamically makes tracking which model generated specific lines of code challenging. According to Mitch Ashley, VP of software lifecycle engineering and AI-native software engineering at The Futurum Group, model choice becomes a runtime dependency rather than a recorded attribute. This makes it difficult for subsequent developers to understand the code's origin, potentially leading to production issues that are hard to trace.

Another significant update is the consolidation of the plugin ecosystem. Agent Plugins 1.0, now generally available, offers uniform functionality across various platforms, including VS Code, Copilot CLI, the GitHub Copilot SDK, and the Copilot app. This change benefits teams who have invested in internal tooling or extensions for a specific platform, as they no longer need to rebuild them for other surfaces.

The Copilot app also introduced side chat functionality, allowing developers to have a separate conversation with the agent without disrupting the main workflow. This feature addresses the challenge of agentic workflows, where agents may interrupt the primary conversation with clarifying questions.

The Copilot CLI received substantial updates in this release. Developers can now manage subagents and monitor their status using a dedicated /tasks command, reducing idle time while waiting for tasks to complete. Queued prompts, commands, and slash commands enable developers to keep multiple instructions ready, even while an agent is currently working on a task.

Additionally, combining the –plan and –mode autopilot flags in headless mode automates both planning and implementation, perfect for CI pipelines or scheduled jobs without requiring manual intervention.

For JetBrains users, two noteworthy additions stand out. Copilot Memory now maintains context across chat sessions, eliminating the need to re-explain project details during each new conversation. Furthermore, JetBrains now supports Ollama as a bring-your-own-key provider, enabling users to leverage local models without relying on external APIs.

Overall, these updates demonstrate GitHub's shift from promoting a single superior model to providing a comprehensive platform for managing multiple AI models effectively. While these changes enhance convenience and control, they also present challenges in terms of traceability and model dependency tracking. Addressing these concerns will be crucial for widespread adoption across development organizations.