---
title: "One year in"
date: "2023-12-20"
summary: "A year after shuttling code between two windows, the model moved into my editor — and the gap I can't shake isn't intelligence, it's what it knows about us"
tags: ["ai","founder","grounding","trust"]
---

A year ago I worked in two windows. One held my code; the other held the model, off in a browser somewhere, and I spent the day shuttling between them like a translator carrying sentences across a border. Now there is no border. The model reads the open file before I've finished describing what I want from it.

That is the year in a single change of address. It moved into the building.

I keep a small private set of problems I throw at every new model. In December the answers were quick and thin. By spring they were clearing bars I'd quietly assumed would hold for a decade. By autumn the labs had stopped feeling like one race down a track and started feeling like weather — a calmer model out of Anthropic that holds a long document in its head, an editor that can finally see a whole project at once, and two weeks ago Google putting its own name on the board.

I have lived through real waves. The way you tell tide from spray is whether the ordinary changes. Mine has. I no longer open the documentation first. I say what I want and read back what arrives.

And still the one gap won't close. The model is brilliant and knows nothing about my studio — my rules, the way we work na míru, the decisions we made in 2019 and never wrote down anywhere. Every conversation it rebuilds that picture from nothing, gets it a little wrong, and hands the guess to whoever asked.

So that is where my head sits at year's end. Not on how clever it can get. On what it knows about us, who owns that knowing, and whether the answer it gives back can be trusted.

Next year I want to build toward that.