Meeting Github Copilot coding agent

I recently upgraded my GitHub Copilot plan to Pro+, which finally allowed me to start using the coding agent in standalone mode. It’s almost like hiring a junior programmer for $39 a month.

At first glance, everything went as expected. I created several issues in my company’s private repositories (yes, you can use your private Copilot in your company’s repos). Copilot began working, and within 10-15 minutes, it started reporting that everything was done and I could begin reviewing its Pull Requests.

That’s where the fun began. My first task was to migrate a CI build from Jenkins to GHA, meaning we needed to port a Jenkinsfile to GitHub’s workflow YAML. The first issue I noticed was that Copilot doesn’t detect failed builds. Second, it can’t read failed build logs. So, when I nudged it with a ‘hey, build failed’ comment, it started peering into the sources and the new build.yml, guessing why the build might have failed. It didn’t find the root cause, but it did generate a new commit with tons of debug logs in build.yml.

I then explicitly told it what the problem with the build was and asked it to revert the last commit.

There were a few more build problems, but when I posted the build logs in the PR review comments, the bot quickly fixed everything. In the end, we merged the PR, and my initial experience is this: you get what you pay for – a junior programmer who constantly needs your attention, but at least you don’t have to work on boilerplate or tech debt tasks.