The Problem: Scaling Beyond a Single OS
I had an image build pipeline that worked fine for a single OS version—but scaling it to support other versions? That was a different story. The logic was too rigid, the structure too tied to a single case. It needed rethinking.
The Collaboration Approach
Rather than dumping the problem on GPT and hoping for a miracle, I treated it like a collaboration. I had the architecture in my head, and GPT became my rubber duck that talked back—with code samples. We went back and forth, re-architecting, re-implementing, and re-deploying. The result: an image builder that scales to multiple OS versions while reusing the service elements each image needs.
Where AI Helped
Speed, mostly—but not just in writing code. The real value was in having a dedicated, always-available resource to bounce ideas off. I could test a thought, get immediate feedback, and see where I might be missing opportunities.
One of those "you know, I hadn't even considered that" moments came when GPT suggested a Makefile that mimicked the CI/CD pipeline for local testing. I hadn't used make in ages, but it reminded me how powerful it can be. Now, almost every repo I own has one to streamline local validation.
Where It Struggled
The CI/CD YAML was a sticking point. GPT could get the rough outline, but the details—runners, permissions, environment-specific steps—needed more guidance. Terraform was a bit better, but still benefited from me feeding in snippets and giving nuanced, targeted instructions.
Just handing over the reins with "go build this" was a fast path to disappointment. But when I treated it like a junior engineer—providing context, examples, and constraints—it was far more effective.
The Takeaway
Used as a partner, GPT sped up a complex re-architecture. It gave me useful suggestions, nudged me toward tools I'd forgotten, and made the work faster without lowering quality. But the magic wasn't in letting it "own" the problem—it was in staying in the loop, guiding the process, and making the final calls.
Next in the Series
What I don't trust AI with (yet)—the parts of infrastructure that still need a human hand on the wheel.
