Set and Forget Doesn't Work with AI

I work with teams that use AI for development and marketing. There's this common belief that you set up a script once, download a skill for Claude Code, and you're done. Create a custom command, set the rules, move on.

That's not how it works.

I have a custom command for creating videos. One file, almost 1,500 lines. It handles scripts, titles, grammar checks, script quality evaluation, studio automation. The first version was 125 lines, created on December 14, 2025. Today it's 1,485 lines with 35 commits over 3 months. That's 12x more. And I still make changes at least once a week.

Because my flow changes. I find things I need to do differently. I realize a step is missing or a check is wrong. Most times I use it, I see something to improve.

I've seen this play out in teams too. One company with 30 people thought one person would set up code guidelines for AI and suddenly everyone could ship PRs. Even non-technical people. Because the skills and rules are there, right?

From my perspective, that can't work. Skills and guidelines are there to speed things up. Fewer repeated mistakes, less back and forth, fewer corrections. But you still need to know what you're doing and where you're heading. You don't just dictate something into a skill file and get good output. The opposite is true. If you want quality, you have to dedicate real time to it.

"Set it once and it works" used to be fine for simple internal tools. Back office stuff. But the moment you properly deploy AI into your company, your product, your processes, you realize your processes are inefficient. And you have to constantly improve them. It's a neverending cycle.

The energy behind this, I think nobody is prepared for. The teams I've worked with aren't prepared to invest the time that's actually needed. They want the magic. The magic is the iteration.