Most journal apps are a text box and a lock icon. You write, it stores the file, and then nothing happens. After a couple of weeks the document is longer and you haven't gotten anything back, so you stop. That's the version of the product we kept running into while we were figuring out Kual Notes.

We didn't think the problem was “people aren't disciplined enough.” The problem was that writing was the whole product. If you did the work of putting something down, the app owed you a return trip — something you wouldn't have noticed while you were in it.

The useful version of that, for us, looks like a specific observation. Sleep and irritability showing up in the same week, again. You only write after 11pm when a certain person is in the entries. The tone drops on Sundays and you never named it. Closer to a good editor than a coach. An editor circles the repeated line. They don't narrate your feelings back to you.

What we wouldn't ship

There's a version of this product that is very easy to demo and pretty bad to live with. We kept a list of things we wouldn't put in, mostly because we'd seen them in other apps and watched ourselves start performing for them:

• A mood score. Once there's a number, you start writing for the number.
• A streak. Missing a week is a week. It shouldn't feel like breaking something.
• A chatbot that asks how you feel in the voice of a wellness post.
• Anything that pretends it can diagnose you.

We're not clinicians. The model is a reader of your own sentences over time, not an authority. If it can't point at something you actually wrote, it shouldn't say it. That sounds obvious until you watch a model generate a warm paragraph that isn't attached to any entry.

Keep the writing page quiet

We argued about how present the model should be while you're typing. The version that lost was the one where it interrupts — a suggestion, a prompt, a little insight in the margin. It turns a journal into a chat, and chat is a different habit. Usually a worse one for this. You start writing for the reply instead of writing the thing.

So the writing surface stayed dumb on purpose. A place to put words. Then a separate moment where the system answers. If the page feels like a dashboard, people perform for it. We wanted the act of writing to stay boring.

There's AI in the stack. It's not the homepage. If Kual only works when the model is clever, we built the wrong product. On a dull day it should still be a journal you can open and read yourself back. The model is how we catch a pattern you'd skip.

We use the same test when someone asks us to put AI in a client product. What comes back to the person who did the labor? If the answer is a chatbot, we're not convinced yet.