My app connected a philosophy essay to a pasta recipe. That was the problem.

Today I watched SolasVine link a note about consciousness to basically every other thing I’d ever said to it. A thread about how the mind rehearses alternate timelines? Connected to my grocery-shaped brainstorm about content pillars. Connected to a plot point in my novel. Connected, I’m fairly sure, to a recipe.

At a loose enough bar, everything relates to everything. Two ideas share a word, a domain, a vibe… sure, draw a line between them. But that’s not insight. That’s noise wearing a trenchcoat.

And it’s a real problem, because connecting your thoughts across conversations is half the pitch for SolasVine. It’s supposed to remember what you said in different conversations and pull it into what you’re thinking now -to notice the thread you forgot you were pulling. So a version of that feature that connects your existential crisis to your to-do list because both technically involve “choices” isn’t a feature. It’s a liability with good branding.

Here’s what I actually did about it.

I stopped asking the app “are these two thoughts related?” because the answer is almost always a weaselly yes, kind of, if you squint really hard. I made it ask a harder question instead: how strongly, and specifically how? A one-to-five scale where most pairs score a one or a two, and a real connection one idea building on another, or flatly contradicting it is rare. Then I made the app harsh. It hides everything below a four now. If it can’t name the specific link, it stays quiet.

The thing that surprised me is that the same disease had shown up somewhere else the same day. SolasVine’s memory had been faithfully holding onto years of me testing half-baked ideas out loud… products I’d shelved, tangents I’d abandoned, whole directions I’d walked away from. And it was pulling them into conversations they had no business being in. It wasn’t broken. It was doing exactly what I built it to do. It was just too generous. Again

So most of today wasn’t building. It was subtraction. Teaching the app what not to surface.

Which is a little funny, because restraint is the thing I’m worst at as a builder. I generate ideas faster than I can ship them. My instinct, always, is to add…another mode, another feature, another clever connection. And it turns out the product inherited the disease from its maker. The most valuable thing I did all day was teach it to hold back.

I think that’s the actual hard part of building a thinking tool, and I didn’t understand it until today. Connection is cheap. Memory is cheap. The value was never in remembering everything or linking everything -it’s in the discrimination. Knowing which one of a thousand possible connections is the one worth putting in front of you. A thinking partner that surfaces everything is just a louder version of your own overwhelmed brain.

I’d rather build the one that knows when to shut up lol

What’s a tool you use that would be better if it just did less?