GitHub Copilot introduces a new experimental feature called Agent mode that operates in a more agentic way. It can debug code, observe results, and make round-by-round adjustments until the assigned task is completed successfully.
In cases where additional terminal commands need to be run, Copilot will prompt the user to run them in rounds before proceeding to edit the code further. Users must manually enable this feature themselves, and once activated, each round of chat prompts must select the Agent mode of operation again, as well as choose the desired model as before.
In addition to the Agent feature, GitHub Copilot also enhances code editing from chat interactions, or Copilot Edits, which take original commands and code to adjust them to different models. Then GitHub predicts code edits more accurately to insert code into files.
The Agentic working approach has gained increasing interest recently. GitHub Copilot showcases the Padawan project where users simply open an issue and instruct Copilot to take on the task. Copilot completes the task and then asks for human inspection. It is expected to be operational this year.
TLDR: GitHub Copilot introduces Agent mode for a more agentic approach, requiring users to interact in rounds and select the Agent mode manually each time while enhancing code editing and predictions for accurate code insertion. It also introduces the Padawan project for user task delegation.
Leave a Comment