← Back to Blog
4 min read

Delete the Form

AIAutomationUX

Nobody has ever enjoyed filling out a form. We tolerate them because for most of computing history there was no other way to get a fact into a database. You had a name, an amount, a date—and the only door into the system was a row of empty fields you had to type into by hand.

But the form was never the point. The point was the record on the other side of it. The fields were just the toll you paid to get there. And lately that toll has become optional.

Say What You Ate

I track what I eat, and for years that meant the grim ritual every food app demands: search a database, scan a barcode, pick a portion size, guess at the items that aren't listed. I quit more than once—not because I stopped caring, but because the friction was higher than the payoff.

So I built something that skips all of it. I say "bacon egg and cheese on a bagel," and it lands in Apple Health with calories, protein, carbs, and fat filled in. Underneath there's a Supabase database, an edge function that adds up the macros, and an iOS Shortcut that syncs the day's entries to Health on its own. I don't experience any of that. I experience saying a sentence and being done.

The database still gets its clean structured row. I just stopped being the one who assembles it by hand.

A Photo Is a Form You Already Filled Out

The same move shows up in my card business, where it saves a great deal more than a few seconds a day. Listing a trading card used to mean typing the player, the year, the set, the grade, and a title into a form—card after card after card. It's the kind of data entry that turns an afternoon into a slog and still finds ways to introduce typos.

Now I take a photo. A vision model reads the card and hands back a finished listing: player, year, set, grade, title, all of it. Listing a card dropped from minutes of typing to seconds of pointing a camera. The card already had every field printed on its face. It was a form I had filled out years ago—I just needed something that could read it.

Receipts work the same way. When a sale confirmation lands in my inbox, a model reads it and pulls out what sold and what it cost, straight into the books. The vendor thinks it sent me a receipt to skim. My accounting thinks it received clean data. Nobody typed a thing.

Loose In, Strict Down

There's a catch, and it's the same one every time you let software read the messy real world instead of a tidy form: the read is a suggestion, not a fact. A vision model can misread a grade. A parser can trip on a receipt layout it has never seen. Treat every read as gospel and you've simply automated the entry of wrong data—which is worse than a slow human, because at least the human notices when something looks off.

So the natural-language part is only half the system. The other half is the half that doesn't trust it. The structured store still has rules: a card has to be a real card in a real set, a day's macros have to add up, a sale has to reconcile against the single source of truth. The sentence or the photo gets you a draft. The rules decide whether that draft is allowed to become a record.

Loose on the way in, strict on the way down. That discipline is the whole reason "just say it" doesn't quietly become "corrupt the database, but faster."

The Human Was the Bridge

What actually changed here isn't that data entry got easier. It's that the human stopped being the translator. For decades the job was to look at the world—a plate of food, a card, a receipt—and hand-convert it into fields a machine could store. That was never work worth a person's attention. It was just the only bridge we had between how people describe things and how computers keep them.

Now the bridge is software, and the person gets to stay on the human side of it: say the sentence, snap the photo, forward the email. The structured record still exists—it has to—but assembling it is no longer my problem.

The record was always the point. The form was just the tax we paid to reach it. I've stopped paying it.