GitHub has added support for several generative AI models in Copilot, including Claude 3.7 Sonnet, Gemini 2.5 Pro, and GPT-4. Developers can choose the model for different tasks depending on their current needs. The updated Copilot features allow switching between models during a session and using different work modes — “ask,” “edit,” and “agent” — for creating, editing, and automating code.
In the video, two developers demonstrate in practice how the same task — creating a simple hotel booking web app — can be implemented using different models and Copilot modes. For example, in “agent” mode, Copilot received a simple instruction, “implement this,” and fully generated the project structure, backend code in Flask, frontend in Vue, and test data. Another example was the automatic creation of technical documentation along with diagrams that could be previewed before saving.
The developers also noted that configuring Copilot with custom instructions allows you to set rules for coding, style, and standards. This helps achieve consistent results regardless of the chosen model and saves time on repeating requirements. Users can adjust the level of detail, comment style, and even avoid unnecessary explanations in AI responses.
