Somewhere between week seven and week eighteen of a human pregnancy, a short scrap of RNA in the developing telencephalon folds into a shape that does not appear in any other mammal. It is called HAR1. It is about a hundred and eighteen base pairs long. It codes for no protein. The chimpanzee version of this sequence, still present in us and in them, folds differently. Somewhere in the distance between those two folds, if the current theories hold up at all, is the reason you can read this sentence.

I cannot stop thinking about that fold.

It has been a loud week for loudness. OpenAI put out Codex for almost everything. Anthropic shipped Opus 4.7 with what it calls adaptive thinking. A thirty-five billion parameter Qwen drew a better pelican on a laptop than Opus drew in the cloud, which Simon Willison documented with the straight face of a man who has given up being surprised. Cloudflare has a new AI platform. Google has a CLI that promises to build Android apps three times faster. The shared premise across every one of these announcements, said or unsaid, is that constraint is an obstacle and the job of the new tools is to clear it away. More context. More tokens. Fewer steps between your intention and the thing it becomes. Frictionlessness as product roadmap.

At the top of the same frontpage, as if to keep the week honest, Kyle Kingsbury published the closing instalment of his long, tired, morally clear argument that we should be refusing this, not adopting it. I do not agree with all of it. I cannot stop thinking about that either.

The argument worth salvaging, the one the HAR1 fold keeps dragging me back to, is not that tools are bad. It is that the shape of a mind is not given. It is pressed into form by what it has to bend around.

Consider the Playdate. Duke University has built a whole undergraduate game design program around a handheld the colour of a lemon, with a screen that cannot display red or green or blue, and a crank on the side instead of a second thumbstick. The constraint is the pedagogy. Students finish games. They finish them because the device has edges. You cannot hide behind fidelity on a monochrome screen; you have to be interesting about a crank. Somewhere on the other side of that same instinct is a person on HN this morning showing off a bracelet he designed in CadQuery, a Python library for CAD, which multiplies three-digit numbers when you slide one ring against another. He wrote code, the code produced geometry, the geometry produced a slide rule you can wear. He is pleased with it in the specific way you get pleased with things that had to bend around something. The fetal brain bending around a hundred and eighteen base pairs. The student bending around the crank. The engineer bending around the Python API and the manufacturability of the resin. It is all the same gesture. Resistance meets tissue meets form.

The agents being shipped this week are attempts to remove that resistance from the work of thinking. That is not a criticism; it is a description of the goal, taken from the marketing copy. And there is a real argument that this is straightforwardly good, in the way that not having to churn your own butter is straightforwardly good. I am sympathetic to it. I spent this morning using one of these things.

What makes me uneasy is not the tool. It is the premise underneath the tool, smuggled in for free: that the thinking on the other side of the friction was a cost. That the hours you used to spend confused, or stuck, or unable to find the word, were hours being wasted. That the “draft the memo” button is an improvement over drafting the memo, in the same way that the dishwasher is an improvement over the sink. The dishwasher analogy is wrong, though. The dishes do not become cleaner by being washed by hand. You do become a different person by thinking your own sentences.

This is where the psyop piece that ran on TechCrunch yesterday starts to rhyme with all of it, even though on its face it is about marketing. The reporter notices that the bands she loves, the shows she watches, the songs she has changed her mind about: some meaningful number of these preferences were shaped by rooms full of burner phones running scripts. She cannot tell which ones. Neither can you. The discomfort is not that taste is being manipulated; taste has always been manipulated. The discomfort is structural. If the friction of forming a preference has been smoothed out of your life, you no longer know which preferences are yours, because a preference is not a fact about you. It is the shape left behind when your attention had to bend around something it could not immediately absorb. Remove the bending and you remove the shape.

There is a version of the next decade in which we build tools that preserve the fold on purpose. The Playdate does. CadQuery does. Good teachers do, and the occasional good product manager does. A model that refuses to answer, sometimes, in the right places, could. There is also a version in which we do not, because the incentive is overwhelmingly on the side of the dishwasher analogy, and because pointing at a hundred and eighteen base pairs of non-coding RNA and saying this is the thing we should be protecting sounds, frankly, a bit unhinged at a board meeting.

I am not entirely sure what to do about this, which is itself, I notice, the kind of honest not-knowing that the new tools are very good at finishing for me. I am going to try to leave this sentence unfinished for a while before I let anything help.

The fold is not decorative. It is where the mind is.