GitHub Copilot is an artificial intelligence (AI) chatbot for coding platforms that is acquiring several new features. The company announced Thursday that it will add a new proxy mode for the chatbot, which can automatically iterate over its generated code and analyze errors. In addition to the Microsoft-owned platform, the platform also offers commonly available copy edits, which was first announced on the Github Universe in October 2024. The company also took off Padawan Project Project Project, the first software engineer (SWE) AI agent for the platform.
Github is introducing new features to Copilot
Just a week after Github added the DeepSeek-R1 AI model to its platform, the company made some improvements to the Copilot AI Chatbot. As detailed in the blog post, the coding platform now introduces a new proxy model to chatbots.
Compared to the usual chatbots, the proxy pattern in GitHub adverbs will be more independent and proactive. It can iterate based on its own code, identify errors and fix them without human user intervention. It also has the ability to analyze runtime errors and can resolve them as well. Additionally, it will suggest terminal commands to the user. The developer will be able to specify tasks, and the proxy mode adverb will also complete all subtasks that may need to complete the main task.
Proxy mode in Copilot can be activated by first downloading the VS code insider and then enabling the proxy mode settings for GitHub Copilot chat. After that, the user will have to go to the Copilot editing panel and then go from edit to proxy through the model selector.
Now, copy editing is usually available in VS code. Using this feature, developers can specify a set of files that require code rewriting or editing, and then use natural language to describe what they want. Copilot edits will understand the commands in the context and then make inline changes to all files. Users can view suggested changes through subsequent queries, accept them and iterate over them.
Github explains that the feature uses a dual-model architecture to improve editing efficiency and accuracy. Users can pick the basic language model from GPT-4O, Claude 3.5 sonnets, and Gemini 2.0 Flash to understand the full context of the editing session and generate initial editing suggestions. The guess decoding endpoint developed internally will then propose these changes in the editor.
Finally, Github also shared Project Pardawan Project Padawan, an independent software engineer (SWE) agent that will be shipped to Copilot later this year. The company said it is possible to assign an AI proxy using any GitHub client in Copilot and will generate a fully tested pull request. Once the task is completed, the chatbot will also assign human reviewers to pull requests and automatically resolve any feedback they leave behind.
According to the post, an AI agent will be provided with a secure cloud sandbox to suit every task it undertakes. Copilot oversees the entire process of adding repositories, setting up environments, analyzing codebases, editing files and building, and performing testing.
