Skip to content

Pick your style: Denkbaar

Book a conversation

Articles

History

Copilot gains a mode that continues working on tasks

·

GitHub presented agent mode for Copilot as a preview in VS Code.

A developer reviews connected files and a loop between an error, a change and a check.
Agent mode could recognise errors and try again. Illustration: Denkbaar.

On 6 February 2025, GitHub presented agent mode for Copilot in VS Code. This preview could follow changes across multiple files, recognise errors and try again. The system could also suggest terminal commands and ask the user to execute them. GitHub explained how it worked. A screen shows three files with an arrow looping back to an earlier step.

In our view, the historical shift lies in that repetition. Assistance did not stop at a code suggestion. The result of a change became part of the next step. This makes the definition of a finished task more important. Hands compare a faulty input field with the repaired version of the same form.

A small or medium-sized business could, for example, ask a builder to fix a fault in a practice form. Describe which behaviour is wrong and which behaviour should work. First have the existing fault demonstrated.

Then ask for evidence that the change fixes that fault and keeps other fields usable. Agree which files and commands belong to the task. Assess the whole result, including the checks. The number of attempts or changed lines is not a useful endpoint for an assignment given to a coding agent.

Sources

  1. GitHub Copilot agent mode (preview)

Topics: devtool