---
title: "Capability got cheap"
date: "2025-04-17"
summary: "Three strong models shipped this month and I forgot which lab made which — when capability turns into a utility, the value moves up the stack to trust, reuse, and memory"
tags: ["ai","models","trust","building"]
---

Three model releases this month and I had to look up which lab shipped which. That tells you everything.

GPT-4.1, o3, o4-mini, Gemini 2.5 — a few weeks ago each of these would have been the headline for a season. Now they arrive, clear my hardest private tests, and I move on by lunch. The thing I waited fifteen years for is becoming a utility. You turn the tap, intelligence comes out, and the price keeps dropping.

I am not complaining. This is what abundance looks like. But I remember when the whole question was whether the model was smart enough to attempt the thing at all. That question is closing. Soon it will be no more interesting than asking whether your electricity is strong enough to run a lamp.

So where does the value go when the smart part is free?

It goes up. To the things capability alone never gives you. Does the answer hold up — can I trust it, trace it, stand behind it in front of a client. Can I reuse what worked last time instead of conjuring it fresh and praying. Does the system know who I am and what my studio actually does, or does it greet me as a stranger every single morning.

None of that is solved by a smarter model. A more brilliant amnesiac is still an amnesiac.

For twenty-five years the hard part of this trade was raw ability — finding it, affording it, squeezing it out of the hardware you had. That era is ending in front of me. The scarce things now are trust, reusable foundations, and memory you own.

The models got cheap. The interesting work just got expensive.