AI Development Tools Get Adjustable Parameters
Microsoft and GitHub are giving developers more control over AI-assisted development tools by introducing adjustable parameters for effort and reasoning. The latest example is GitHub's new code-review control, which allows developers to choose between 'Lite' and 'Balanced' review efforts. Lite provides faster, targeted feedback on routine changes, while Balanced routes a pull request to a higher-reasoning model for longer analysis.
The thinking effort dial in Visual Studio also exposes an adjustable parameter that lets developers tell supported models how much reasoning to apply to a request. The available levels vary by model and include Low, Medium, High, Extra High, and Max. Developers can configure the level from Visual Studio's model management interface, where thinking effort appears alongside information including model capabilities, context size, and cost.
These controls are part of a wider set of changes that increasingly expose parameters behind AI interfaces. Recent Microsoft and GitHub additions include reasoning levels for Copilot cloud agent, reasoning and context controls across Copilot, and a unified model-tuning control in VS Code.