Github has announced numerous updates to Copilot, and also gives you a glimpse into the future of more agents for AI-powered pair programmers.
Notable updates include a feature called Vision from Copilot. This allows users to attach screenshots, photos, or diagrams to the chat.
So, for example, someone from the marketing team could take a screenshot of a web page and explain some of the changes needed to that page. Instead of requesting such changes via a text prompt, it is now possible to upload images and ask Copilot to implement the changes as shown in the file.
It will be available as an extension to VS code from October, and it’s worth noting that Microsoft has confirmed in its app description that it will ultimately be deprecated in favor of native features within Github Copilot Chat. It deserves. Previously, users had to have their own API keys.
GitHub introduces a new feature called “Next Edit Prostations” as part of Copilot’s wider code completion feature.
Until now, Github Copilot, the editor of VS Code, worked from the cursor position, but now I look into other recent edits and predict what I want to do next. Essentially, it uses a richer context queue to automatically identify and suggest what the next edit should be.
If the developer wants to accept the suggestion, they can either press the tab key or reject it and press ESC.
The future of Github agents
Github debuted Copilot Edits last November, serving as a new way to perform multi-file editing using natural language prompts. This feature currently packs a new “agent mode” that graduates from general availability and identifies all files related to the changes the developer is about to change. Must be applied to.
“Adjunct officer 1738909714 Github CEO Thomas Dohmke told TechCrunch:
In the long run, the goal is to apply such “agent mode” to other aspects of Copilot, allowing for greater automation across a wider range of interlinked tasks. This is why GitHub bullied the new initiative they call Project Padawan. This is essentially a SWE (Software Engineering) agent that can handle the entire task independently under the directions of the developers who assign problems to Copilot.
Dohmke gave no indication when this could be prime time ready. We simply focused on working with our partners and communities to improve it.
“We are also looking for partners who are building agents to integrate into that flow. We work with our customers and provide feedback because we know that it’s not perfect at this stage. “Dohmke says.
TechCrunch has a newsletter focused on AI! Sign up here to get it every Wednesday in your inbox.
Source link
