There's a concept in software development called refactoring.

The name sounds technical, but the idea is simple. Code develops unhealthy shapes over time. Files get too large. Modules know too much about each other. One thing changes and three other things break. Experienced developers learned to recognize these warning signs and fix them before the system collapsed under its own weight.

It became a craft skill. One of the better ones, honestly.

AI kind of made it feel like a lost art.. because when the model is generating your code, you stop making those structural decisions yourself. The machine handles the syntax. You just describe what you want.

Here's the thing though. The skill didn't disappear. The layer it applies to just shifted.

When you start building multi-step AI workflows.. chained agents, orchestration layers, skill files that drive LLM behavior.. you're not writing code anymore. But you're absolutely making the same architectural decisions. And the same bad habits that destroy codebases are destroying AI workflows right now, in the wild, on real builds.

Skills that do too many things at once. Prompts that repeat the same logic in five different places. One component that knows too much about another one, so pulling a thread in the wrong place breaks the whole chain.

The symptoms are identical. The names just changed.

So here's what a 25-year software developer does when he looks at an AI workflow:

He asks if each skill has a single clear responsibility. Can you describe what it does in one sentence without using "and"?

Look for duplication. If the same instruction or context is showing up in multiple places, that's a maintenance problem waiting to happen.

Look for coupling. Do your skills depend on specific implementation details of other skills? If skill A has to know how skill B works internally, you've got a problem.

Make sure skills are composable. A skill should drop into different workflows without needing to be rewritten. If it's only usable in one specific context, it's not really a skill yet.

These aren't new ideas. They're principles that have held up across every programming paradigm shift for decades. They hold up here too.

The people building AI workflows right now are making every mistake developers made in the early days of object-oriented programming. The workflows feel fast. They feel powerful. And underneath, they're fragile in ways that won't reveal themselves until they break in front of someone who matters.

Worth thinking about what your workflow actually looks like at the structural level.

If you're building in this space and want to talk shop.. come hang out: www.facebook.com/groups/appsandfunnels/

🚀

​- James