AI: A Year Later
- #ai
- #workflow
- #architecture
Living and working in the age of AI is a bit different. AI is inevitable, and we can’t fight it, so we should embrace it. It doesn’t end there. Even after we are willing to embrace it, we need to realize that it evolves much faster than we ever imagined, forcing us to change ourselves. Like it or not, we need to adapt how we use and collaborate with it. We’re entering a new era where we must revisit our way of working periodically, otherwise we can’t keep up.
I’ve been working with AI, using pair programming to refactor an old, stale project from design to a full product in just two days. It’s true that I’m not starting from scratch this time, the infrastructure is already settled. The thing is, that infrastructure was also created by pair programming with AI a year ago. The funny thing is, it feels different. One year apart, and AI has become so much more reliable.
But it’s not without a catch. I feel dizzy jumping back and forth between design layouts and terminals. It’s not easy to keep up with AI when it works faster than I can process. Even with the experience I’ve gathered so far, I feel that I can easily lose track during pairing. And I’m aware that if the initial work is wrong, AI will treat it as the rule to follow, drifting further off course until the project stops being maintainable.
So I took a step back to things that used to be ignored when I worked by myself manually, which are linters, architecture DSLs, tests, and golden tests. CI/CD, linting, and testing (including golden tests) feel much more useful now. They let me focus on ADRs, architecture, and rules, and then the code implementation follows and gets checked automatically. For the UI, I rely on golden tests to catch any drift during UI refactoring produced by agentic AI. This time, I must employ as many test frameworks as available to cover each side of the system I build.
One thing I’ve learned from this process is that we should build big projects in smaller chunks, loosely coupled packages, and put a harness around each of them. Only by doing that can we get the best out of it while keep staying in the same pace to our pairing partner.