Logo
WritingReadingProjects

Breaking the Curse

We know what's hard, so we stop asking for it.


We former developers have a curse. We know what’s hard, so we stop asking for it.

We know how long things take. We know what’s easy and what’s hard. We know which feature takes an afternoon and which one eats a quarter.

That knowledge used to be the job, and it still matters. But now AI writes the code, and if we’re not careful, the knowledge decides what we even ask for.

Same tone, different planet

Watch someone who doesn’t code use Claude. They sit down and ask it to do some magic with a spreadsheet. Easy.

Then a marketer asks for the next thing, in the same casual tone, with no details:

“Show me how much revenue each customer brought in, counting everyone they referred, and everyone those people referred.”

To them, it’s the same kind of request. Another sum. Another column.

What we see

We read that sentence and see a graph. Nobody knows how deep the chain goes, so a single join won’t cut it. We have to keep walking it until it ends. Messy data can make referrals loop back on themselves, so we have to track who we’ve already counted, or we loop forever and count people twice. And on a big customer table, the naive version crawls.

A partygoer stands alone in a corner thinking, “They don’t know how to traverse a cyclic graph.”

I won’t get into complexity theory. The point is that we know these two requests are vastly different.

So we do one of three things:

  • We don’t ask.
  • We scope it down to direct referrals “for now”.
  • We start planning before we type a single prompt.

The marketer just asks. The AI does it and handles the details.

Why it’s a curse

We hold ourselves back. Not because the AI said no, but because we assumed it would.

We filter our own ideas through what used to be hard for a person to build. People who don’t know any better skip that filter. They ask for everything, and the AI handles it.

We also chase quality. Tabs vs. spaces, Redux vs. React vs. Vue, BFS vs. DFS. It’s the same instinct every builder has. Build a house on sand and it goes down eventually. Build it on solid brick and it lasts. The question is who’s willing to pay for the brick.

Right now someone just took the lid off the pot, and everyone is shipping apps at the speed of light. Developers were never known for being fast. We have to adapt, even if what we ship won’t stand forever on its own.

The knowledge isn’t the problem. Letting it shrink our asks before we even try is.

Breaking it

Ask first. Ask like the marketer: the whole thing, in plain words, without scoping it down in your head.

Let the AI hit the wall before you assume it will. If it does hit one, that’s when the curse turns useful again. You know why it failed, and you know what to ask next.