Copilot X is moving beyond code suggestions. Microsoft and Github have updated the coding assistant to GPT-4 and added features that let developers ask questions about code, consult software documentation, generate pull requests, and work with command-line instructions.
Chat brings questions into the coding process
The central addition is Copilot Chat, which lets a developer discuss a specific segment of code with GPT-4. A user can ask for help understanding or changing that code, and the chatbot can identify errors, suggest improvements, or make changes based on natural language input.
This gives developers a way to move between reading code and asking for help about it within the assistant. Instead of relying only on code autocompletion, Copilot X can respond to a request about what a particular section does or how it might be improved. Microsoft describes the experience as similar to ChatGPT, and says voice input will also be available.
The voice option extends the same conversational approach beyond typing. The source describes it as a way to code without a keyboard, while the broader point is that assistance can take the form of an exchange rather than a suggested line of code.
Documentation and pull requests add new tasks
Copilot X also offers chat responses based on software documentation. The feature uses GPT-4's longer context option to make documentation information available through conversation. React, Azure Docs, and MDN documentation are available initially.
That gives developers a way to ask about material from those documentation sources through chat. It puts documentation help alongside questions about code, though the available sources are limited to the named selections at the outset.
Another addition can automatically generate pull requests using GPT-4. Developers can review and approve those proposals. That review step leaves developers responsible for deciding whether a suggested change should proceed, even as the assistant helps produce it.
Command-line help broadens Copilot's reach
GPT-4 is also used in Copilot's command-line interface. Github says it can compile commands and loops, including what it calls “obscure find flags.” This takes the assistant into a different part of development work: assembling command-line instructions rather than completing code in an editor.
Together, the chat, documentation, pull request, and command-line features point toward a broader role for Copilot X. The assistant is described as supporting developers across the development lifecycle, rather than focusing only on writing code. Each feature addresses a different kind of task, from understanding an existing code segment to preparing a proposed change.
That broader ambition is significant to the product's positioning, but the source presents these capabilities as new features rather than evidence that all development work can be automated. Pull requests still require review and approval, and access to the new features is limited to a technical preview or waiting list.
GPT-4 replaces the earlier model approach
The original Copilot relied on OpenAI's Codex model, which Github described as a code-optimized derivative of GPT-3. Copilot X's move to GPT-4 marks a change in the model behind the assistant: GPT-4 is described as a general-purpose language model that has also learned to code.
Github says it is seeing “significant gains in logical reasoning and code generation” with GPT-4. The source connects that development with OpenAI's announcement that Codex will be discontinued, while noting that researchers can still access it.
For developers, the change is presented as part of a shift from isolated autocomplete toward assistance across more stages of work. Copilot X's features put that idea into practice through conversation, documentation answers, command construction, and pull request generation. For now, those features remain available through a technical preview or waiting list.