GitHub has shared some exciting updates for its AI tool called Copilot, which acts like a helpful coding assistant for developers. They also hinted at a future where this AI can take on more tasks independently.
One of the standout new features is called “Vision for Copilot.” This allows users to attach images—like screenshots, photos, or diagrams—to their chats with Copilot. When a user uploads an image, Copilot can understand it and generate the necessary code and descriptions to make changes based on what’s shown in the image.
For instance, if someone from a marketing team takes a screenshot of a webpage and wants to suggest some changes, they can simply upload that image and ask Copilot to make those changes. This is much easier than typing out detailed instructions. This feature has been available as an add-on in a code editor called VS Code since October, but now it’s becoming a built-in part of GitHub Copilot Chat, making it more accessible. Previously, users needed a special API key to use it.
Read Also: Who Is Liang Wenfeng? Founder Of DeepSeek With A Background In AI Investing
Another new feature is called “next edit suggestions.” In the past, Copilot would only look at where the cursor was in the code to suggest the next step. Now, it can consider recent changes made in the code to better predict what the developer might want to do next. This means that Copilot can provide smarter suggestions based on what you’ve just done, making the coding process smoother. Developers can easily accept these suggestions by pressing the Tab key or reject them by pressing Esc.
GitHub also introduced a feature called Copilot Edits last November, which allows developers to make changes across multiple files using simple language commands. This feature is now fully available and comes with a new “agent mode.” This mode helps Copilot identify all the files that are relevant to the changes a developer wants to make, instead of the developer having to pick out each file manually.
According to GitHub’s CEO, Thomas Dohmke, Copilot is getting better at understanding what a developer wants with their requests and can do more of the work to figure out how to fulfill those requests. Looking ahead, GitHub aims to expand this “agent mode” to cover more tasks, which would allow for greater automation in coding.
Additionally, GitHub has teased a new project called Project Padawan. This initiative envisions a software engineering agent that can handle entire tasks on its own, while still being guided by a developer who assigns it specific issues to work on. Although there’s no timeline for when this will be ready for widespread use, Dohmke mentioned that GitHub is collaborating with partners and the community to refine this tool. He emphasized that they are looking for feedback as they develop it, acknowledging that it won’t be perfect right away.
