Along with a sneak peek at a more agentic future for its AI-powered pair coder, GitHub has revealed a number of enhancements for Copilot.
Vision (preview for the public) which one may now deal with photos directly in Copilot Chat by attaching them. Send Copilot screenshots of mistakes, and it will decipher the picture and fix the problem. You may also submit fresh design mockups, and Vision will assist you in making them a reality.
The Vision for Copilot feature, which lets users attach a screenshot, image, or diagram to a conversation and has Copilot create the interface, code, and alt text to make it come to life, is one of the noteworthy changes.
A member of the marketing team may, for instance, take a screenshot of a webpage and use it to demonstrate certain modifications they would like to see done. Uploading a picture and asking Copilot to make the adjustments as specified in the file is now an option instead of using text prompts to request such changes.
Notably, Microsoft confirmed in the app’s description that this will soon be retired in favour of a native feature within GitHub Copilot Chat, even though it has been accessible as an extension in VS Code since October. In the past, the user also needed to have their own API key.
A new feature called “next edit suggestions” is also being introduced by GitHub as part of Copilot’s expanded code-complete capabilities.
GitHub Copilot used to operate from the cursor location in the VS Code editor, but now it looks at other recent updates to anticipate what you might want to do next. In essence, it uses deeper contextual clues to automatically determine and recommend the next update.
The developer can use the Tab key to accept the ideas or Esc to reject them.
The agentic future of GitHub, last November, also introduced Copilot modifications, a new method for performing multi-file modifications with natural language prompts. In addition to being moving toward universal availability, this feature includes a new “agent mode” that, instead of requiring developers to explicitly choose which files the changes should affect, finds all the files that are pertinent to the changes they are attempting to make.
Thomas Dohmke, the CEO of GitHub, told TechCrunch that Copilot “now does more of the work to figure out what you had in mind with your original request, and then tries to solve that.”
The long-term objective is for these “agent modes” to be applicable to additional Copilot features, allowing for more automation of a wider variety of interconnected activities. For this reason, GitHub has also hinted at a new project it’s calling Project Padawan. In essence, it’s a software engineering (SWE) agent that can manage full tasks on its own, guided by a developer who gives Copilot concerns.
Dohmke just stated that it will be working with partners and the community to develop it, without providing any indication as to when this could be ready for prime time.
Because we know it won’t be flawless at this point, Dohmke stated, “We’re looking for partners that are also building agents, to integrate into that flow, and for customers to work with us and give us feedback.”
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
