Last week I was building a monitoring dashboard with a client to track their public companies.

The first pass was embarrassing. Missing Glassdoor entirely. No Reddit. No X data worth looking at. Stock prices were flat-out wrong. It was the kind of work that makes you think the AI wasn't really paying attention.

We spent an hour refining it.

AI *can* create things, but can it create what *you* are looking for?

In the case of building a news feed for the public companies, our brains know what's noise and what's relevant. For example, I see someone on X.com cramming five tickers into one post and I instantly know it's junk.

How do I teach the system that?

We set rules. X accounts get ONE ticker mention only. 10k follower minimum. Suddenly the signal got cleaner.

But that instruction isn't obvious when you're building your first iteration. It evolves over time with practice and refining what "great" looks like.

There's a massive upfront cost to make the system think like you do. The AI doesn't come with your judgment built in. You have to install it yourself.

It's like hiring an intern. A really competent one who works 24/7. But still an intern. You're constantly checking the work, explaining context, catching the stuff that slipped through. You can't really fire this one or give it less responsibility. You have to keep supervising.

Most people imagine AI automation as set-it-and-forget-it. What they don't see is the calibration phase. Teaching it what matters. Filtering for signal. Defining the edge cases that will haunt you if you don't.

If you're building with AI, the first pass is always half-baked. Budget for the refinement work. The conversations, the rule-building, the iterations.

Alex

PS: Reply and tell me what you're building.