GitHub Copilot Generates Code from Images

GitHub has unveiled a series of significant updates to Copilot, its AI-powered pair programmer, offering a glimpse into a more autonomous future for software development. These enhancements include Vision for Copilot, improved code completion, and agentic capabilities aimed at streamlining workflows and boosting developer productivity.

Vision: Code Smarter with Contextual Understanding

One of the most notable updates is Vision for Copilot, which enables users to attach images, screenshots, or diagrams to chats. Copilot can then generate interfaces, code, and alt text to bring these visuals to life. For instance, a marketing team member can upload a screenshot of a webpage and illustrate desired changes, and Copilot will implement those changes as indicated in the file. This feature, previously available as an extension in VS Code, is now natively integrated into GitHub Copilot Chat.

Microsoft has confirmed it will eventually be deprecated in favor of a native feature within GitHub Copilot Chat.

Enhanced Code Completion with "Next Edit Suggestions"

GitHub is also introducing "next edit suggestions" to augment Copilot’s code-complete functionality. Instead of only analyzing the cursor position, Copilot will now consider recent edits to predict the developer's next move. By leveraging richer contextual cues, Copilot can automatically identify and suggest subsequent edits. Developers can accept suggestions by hitting the Tab key or reject them with Esc.

Agentic Future: Copilot Edits Graduates to General Availability

Copilot Edits, initially debuted in November, offers a new method for conducting multi-file edits using natural language prompts. This feature has now reached general availability and includes a new "agent mode" that identifies all relevant files for a given change. Instead of developers manually selecting files, Copilot can now automatically determine where changes should apply.