# Outkicking Your Coverage

But there is a moment where every AI project gets tested: when you need to actually ship a feature or a tool. I like to say that you can *vibe* your way to almost anything — the real challenge shows up when you try to share that project with other people. That tension has pushed my own workflow through a few distinct stages, and I wanted to write them down.

## Stage one: spec-based development

At first, I could build incredible things, but they always felt a little incomplete — not quite right. That pushed me toward **spec-based development**: writing a complete specification of the feature or tool *before* the agent writes any code. The agent grills me on the ins and outs of the feature, forcing me to think through the edge cases I would otherwise have discovered the hard way. A good spec front-loads the thinking, and it is one of the highest-leverage habits you can bring into software engineering with AI.

## Stage two: outkicking your coverage

Then you hit the next wall. You are building so much, so fast, that you cannot possibly test all of it.

Right now I am working on an integration between two projects — adding a little to one and a lot to the other to get the seam between them right. This is where I have started to outkick my coverage. More thought leads to more code, which leads to more testing, and I cannot keep up. Automated tests help guard against regressions, but they do not absolve you of the work: you still have to *learn* the software you built.

That is the part that is easy to forget. You have to use the thing. You have to test it and live with it. And right about the time you get comfortable, you understand it well enough to want to rework the whole thing.

## What I am taking away

I am noting this journey for a few reasons:

- **Using AI is not a license to throw out software engineering best practices.** The fundamentals matter more, not less.
- **If you think about your project less, your product will be less than you hoped.** The quality of your output tracks the quality of your attention.
- **Architecture-first development still wins.** Top-down design has not gone anywhere. Thinking about architecture up front is what keeps a fast-moving project from collapsing under its own weight.

So that is where I am headed next: hunting for architectural and integration patterns I can bring into my projects — so I can keep building fast without outkicking my coverage.