Agentic Coding On Your Phone Is Actually Good(?!)

I had (another) “Wow” moment with AI at the start of the year. I found myself bored on a train to Cork and decided to give the GitHub Copilot agent on my phone a try. I didn’t expect to be very productive — or, in fact, get anything done — since working on a phone has never been a positive experience for me. But I was quickly blown away by how good the results were and how enjoyable the whole thing was.

I started by letting Copilot do some simple dependency upgrades, but quickly moved on to more involved tasks: Setting up Dependabot, improving the CI setup, writing tests for previously untested code (to give me confidence for future Dependabot updates), and tackling some major dependency upgrades I had been putting off that required dealing with breaking changes. I found myself running multiple tasks in parallel — constantly either reviewing the AI’s code (which admittedly would have been nicer on a big screen), giving Copilot instructions for changes via PR comments, or dispatching new tasks.

By the end of the train ride, I had merged 10+ PRs and made some significant progress on my side project, all without having typed a single line of code myself. I also had a lot of fun doing it. It required no preparation (like bringing a laptop on my weekend trip) and no commitment (I could just put the phone down and continue another time).

This blog post is mostly an invitation: If you have some time on your hands while travelling and want to make some progress on your side project, give GitHub Copilot agents on your phone a try! For me it was a genuine Wow moment, and I think maintaining side projects just became significantly easier.

Note: This article was originally written during my employment at TNG and is republished here with permission. The opinions are my own and TNG does not endorse this publication.