Two years of AI discourse has been dominated by prompt engineering. Craft a better question, get a better answer. There's something true in it. But it's not where the real leverage is.
What actually moved the needle wasn't how I phrased questions. It was the context that existed before any question was asked. The more accurately the AI knew who I was, what I was working on, what I already knew — the better the output. Not because my prompts improved. Because the environment did.
This distinction matters enough to name. A prompt engineer optimizes the question. An AI Environmentalist designs the conditions in which good thinking grows. An instruction tells the AI what to do right now. An environment tells it who you are, how you think, what you're working on, and what you value — persistently, across every session, without re-explanation.
I built this on Obsidian and Claude Code. The vault is organized around states, not topics. Notes move through stages of development. There's a governing document that loads at every session start — not a config file, a teaching document. Memory persists across sessions in two layers. The AI gets smarter about my life over time. Not because the model improved. Because the environment contains more accurate information.
The whole thing runs locally. My notes are markdown files on my machine. No proprietary format, no subscription holding my data hostage, no platform that changes its terms. The system is mine and can remain that way. That's not a technical footnote — it's the point.
Before building this: a blog started four years ago. Struggled weekly, published 12 posts, stopped. Since: 82 daily posts, and I take pleasure in writing them. Not because I became more disciplined. The vault made the ideas visible. Half-formed thoughts from years ago, sitting in forgotten notes, now connectable to something written last week. The system didn't generate those ideas. It surfaced them, and showed me how they fit together.
The deeper surprise was that it revealed a thinking method that had always been mine. The patterns I return to, the instinct to connect things across domains — it was always there. I couldn't see it until something held it up. That's what an environment does. It doesn't add capability. It makes visible what's already there.
You don't need my setup. You need your own.
Originally published at: https://www.asymmetricalcollage.com/articles/2026-04-05-the-environment-not-the-prompt/
What actually moved the needle wasn't how I phrased questions. It was the context that existed before any question was asked. The more accurately the AI knew who I was, what I was working on, what I already knew — the better the output. Not because my prompts improved. Because the environment did.
This distinction matters enough to name. A prompt engineer optimizes the question. An AI Environmentalist designs the conditions in which good thinking grows. An instruction tells the AI what to do right now. An environment tells it who you are, how you think, what you're working on, and what you value — persistently, across every session, without re-explanation.
I built this on Obsidian and Claude Code. The vault is organized around states, not topics. Notes move through stages of development. There's a governing document that loads at every session start — not a config file, a teaching document. Memory persists across sessions in two layers. The AI gets smarter about my life over time. Not because the model improved. Because the environment contains more accurate information.
The whole thing runs locally. My notes are markdown files on my machine. No proprietary format, no subscription holding my data hostage, no platform that changes its terms. The system is mine and can remain that way. That's not a technical footnote — it's the point.
Before building this: a blog started four years ago. Struggled weekly, published 12 posts, stopped. Since: 82 daily posts, and I take pleasure in writing them. Not because I became more disciplined. The vault made the ideas visible. Half-formed thoughts from years ago, sitting in forgotten notes, now connectable to something written last week. The system didn't generate those ideas. It surfaced them, and showed me how they fit together.
The deeper surprise was that it revealed a thinking method that had always been mine. The patterns I return to, the instinct to connect things across domains — it was always there. I couldn't see it until something held it up. That's what an environment does. It doesn't add capability. It makes visible what's already there.
You don't need my setup. You need your own.
Originally published at: https://www.asymmetricalcollage.com/articles/2026-04-05-the-environment-not-the-prompt/