Hi everyone!
GitHub's Copilot just got a major upgrade: a new Coding Agent. The idea is you can now assign GitHub issues directly to Copilot, and it’ll autonomously work on tasks like bug fixes, new features, or writing docs, then submit a PR for review.
So, instead of just getting code suggestions, you now have an AI teammate that takes on coding assignments. It operates in the background using GitHub Actions, analyzes your codebase, and you can see its progress and reasoning through logs. It even gets visual context from images in issues.
Beyond its core coding, it can connect to external tools via MCP. GitHub has also built in important security guardrails, like requiring human approval on PRs and respecting existing branch rules.
This essentially offers developers an AI assistant to handle those routine or time-consuming coding chores, freeing them up for more complex problem-solving. It's available now for Copilot Enterprise and Pro+ users.
Coding agents are awesome! The github copilot agent with sonnet 3.7 in Code is as good as they come! And most of the code is really good. But sometimes when it solves something it also accidentally deletes 900 lines of code. But it usually happens in files that are very long, or when the codebase is not well documented. So a trick is to modularize the code base, and add comments in the code. However, many files also becomes tedious to manage with bots, its a balancing act. Im not sure if im ready to assign tasks to an autonomous agent tho. Supervision is really key. Usually takes a few rounds of chatting before a solution to a problem is found. And without the back and forth, the solution often derail. One thing I do find interesting right now is to fire up 3 agents. and have them work on separate branches on separate problems. Just start 3 windows in VSCode.
