WHAT I'M WORKING ON
Embracing AI Agents
Yes, I know, I'm late to the party! But am I really?
I started working on fundamental improvements to RocketSim, pretty much the only Xcode project I'm currently working on. I implemented more tests, migrated to Swift 6 & Strict Concurrency, and enabled CI to run all tests on pull requests.
There has been a vision behind this: I knew it would be important to onboard AI agents. Yet, I did not want to let AI do anything without it being able to run tests. On top of that, being able to verify all tests succeed on a PR is helpful for both AI agent code and other contributors.
With that fundament into place, I was finally able to focus on adding AI agents to my repository. I'm currently only focusing on Codex, but this is simply a matter of focus—using too many tools at once can distract me from my goals.
The result is a list of Codex tasks that I triggered within an hour while working from a coffee place. I was impressed! Sure, not everything was worth merging or even continueing, but I still fixed quite a few bugs in a much shorter time.
This was also a moment where I realized how important it is to get started with these tools. We're becoming code mentors, code reviewers, much more than developers. It's a skill to coach an agent, to provide proper instructions.
Some of the tasks in the above image didn't result in a PR. For me, the code changes were either of low quality or too many lines of code. I'm still practicing and getting better, but I feel like AI agents are currently most effective for me when focusing on smaller tasks, smaller code to review.
I hear you thinking: AI can do big tasks!
It sure can, but I want to build up momentum gracefully. Learn along the way, get better along the way.