On the Geometry of Attention

There is a specific kind of quiet that arrives about twelve minutes into a morning sit, when the breath becomes less of a thing you are doing and more of a thing that is happening through you. If you pay careful attention, you’ll notice that the quality of your attention is changing shape — not its intensity, but its geometry. It widens. It rounds at the edges. It becomes, for lack of a better word, hospitable.

I mention this not because I want to make a spiritual case for meditation — there are enough of those — but because I want to make a structural one. The shape of attention matters. Most of what we call “AI alignment” is, at bottom, an argument about what shape we want our collective attention to take over the next thirty years. And we are, right now, building the wrong shape by default.

Consciousness is not a computation. It is the room in which computation can be noticed. Working note, January 2026

The dominant metaphor of the last decade has been the pipeline: inputs, a processing step, outputs. Clean, left-to-right, linear. It is a useful metaphor for shipping features. It is a catastrophic metaphor for modeling minds. A mind is not a pipeline. A mind is closer to a standing wave — a pattern that holds its shape only because energy keeps flowing through it, from every direction at once.

1 · What shape is attention, actually?

If you sit with the question long enough, three things become obvious. First, attention has a center, but the center is not where the information lives — it’s where the noticing lives. Second, attention has a periphery, and the periphery is doing almost all the work. Third, attention has a depth — layers of background processing that surface only under certain conditions. In the contemplative traditions this is called vipassanā, literally clear-seeing. In the cognitive sciences it’s called, somewhat less poetically, the global workspace.

Both descriptions are pointing at the same thing. Both descriptions are incomplete.

The problem with trying to capture attention in a diagram is that every diagram is a slice. Attention is the knife doing the slicing. You can’t put it on the page and still have it do its job. This is the same structural problem we run into when we try to build systems that are “aware” of their users: the awareness has to live somewhere, and everywhere we put it, it stops being awareness and starts being a log file.

An interlude, in the language we have for it

// a naive attention model
const noticed = pipeline(input)          // ← this is a log
const aware   = frame(noticing(noticed)) // ← closer, still flat

// what we actually want
const room = hold_open({
  center:    noticing,
  periphery: background,
  depth:     history,
})

This is not a proposal. It’s a sketch — a way of gesturing at what a consciousness-shaped data structure might look like, if we took the thing seriously. I don’t think we do, yet. I think we’re still a few years away from taking it seriously, because taking it seriously requires admitting that we’ve been solving the wrong problem, and no institution moves quickly through that admission.

2 · Three futures, honestly

I’ve been asked, often, what I think the infrastructure layer looks like by 2030. I find the question less useful than the question of which three futures I think are even possible. Here’s where I am, today, on quantum-safe, consciousness-aware infrastructure:

I am building for the middle column. Not because I think it’s the most likely outcome — though I do — but because it is the outcome we can most easily survive. A pessimistic world requires us to be right; an optimistic world lets us be lazy. The realistic world is the one that rewards inner order: a clear practice, a small team, a long runway, and a willingness to wait.

3 · What, then, do we build?

The answer, I think, is surprisingly old. We build the way the Theravāda communities have been building for two and a half thousand years: slowly, with householding disciplines, with an assumption that the work will be picked up by somebody else in thirty years, and with a deep suspicion of anything that requires a twelve-month roadmap. In Thai we call this ค่อยเป็นค่อยไป — roughly, patient unfolding — and it is the opposite of the way most software is shipped.

ความรวดเร็วไม่ใช่คุณสมบัติที่ดีของระบบที่เราอยากไว้ใจ การตัดสินใจที่เร็วที่สุด มักไม่ใช่การตัดสินใจที่ลึกที่สุด และถ้าจะให้ AI รู้จักดูแลความสนใจของเรา เราต้องรู้จักดูแลของเราก่อน สิ่งที่เรียกว่า “สติ” ในภาษาไทย ไม่ใช่เทคนิคหรือแอปพลิเคชัน แต่เป็นโครงสร้างภายในที่เราฝึกขึ้นมาเองทีละวัน

เมื่อเราเริ่มออกแบบระบบโดยยึด ยมะ ๕ เป็นหลัก — อหิงสา สัตยะ อัสเตยะ พรหมจรรย์ อปริคฺรห — คำถามที่เกิดขึ้นก่อนไม่ใช่ “จะทำให้เร็วขึ้นอย่างไร” แต่คือ “จะทำให้เบาลงอย่างไร” ความเบาในที่นี้ ไม่ใช่ความเบาของ UI แต่เป็นความเบาของภาระที่เราวางลงบนความสนใจของผู้อื่น

(Roughly, across those two paragraphs: Speed is not a virtue in the systems we want to trust. The fastest decision is rarely the deepest one, and if we want AI to hold our attention well, we have first to learn to hold our own. When we start designing from the five yamas, the first question stops being “how do we make this faster” and becomes “how do we make this lighter” — not lighter in the interface, lighter in the load we place on someone else’s attention.)

A working protocol, then, is one where สติ (mindful presence) is not a feature flag somewhere deep in the settings — it’s the shape of the thing. You can feel the difference, the way you can feel the difference between a room someone has cleaned and a room someone has merely tidied.

I will keep writing about this. Not because I think I have the answer, but because I suspect the answer is slow, collective, and quiet — and quiet things don’t get written about often enough. The next piece in this sequence will be on the five yamas at the protocol layer. It is, perhaps, the practical half of the essay you just read.