The Understory
The understory is the layer beneath the canopy: shade-tolerant, slower, and the reason the forest holds together. Ours is eleven pieces long. This is all of it in one go, for anyone who would rather read the argument than the table of contents.
One idea, and everything else is a consequence
Water flows one way. Whoever sits upstream decides what arrives downstream, and you do not negotiate with upstream. You receive it.
That is the quiet arrangement in most software. The vendor is upstream: they decide when the update lands, when the terms change, when the redesign moves your furniture, when the notification arrives. You live downstream of the product and you get its weather.
We build it the other way around. The person is the source. Their writing, their numbers, their intent, their attention. The tool sits downstream, drinking from what they make, and its job is to receive and render and never to dam.
Anyone who has maintained open source already has the word for this. Upstream is the project you track, patch against, defer to. So, plainly: in our software you are the upstream project. Changes flow from your needs down into the tool, and not from our roadmap up into your life.
Two rules run in that channel, and everything below is one of them applied to something.
What you make is yours to keep, and it crosses to anyone else, us included, only by an explicit act of yours. Your attention begins in your control, and is granted on your terms.
Which decides who pays us
Whoever pays is upstream of the roadmap. Advertising puts an advertiser there. Engagement dashboards put an investor there. Free puts a data broker there.
A person paying a fair price for a good tool is the only arrangement in which the river runs the direction we just promised. That is not a pricing strategy. It is the riverbed.
What it costs the tool
Most software is loud on purpose. The badge that will not clear, the streak you will break, the red dot bleeding into your evening: those are not accidents of design, they are the design. We build the quiet ones. A tool of ours waits. You open it when you have something to do, you do the thing, you close it, and it does not hold the door. Calm is not the absence of features; it is one we ship, and it costs us the cheapest trick in the book, because we will never manufacture a reason for you to come back.
A hammer does not get a quarterly update. It was finished the day it was good and has been quietly good ever since. Somewhere we all decided software was different, that a tool which stopped changing must be dying. We think a tool can be done. Not abandoned and not starved: done. Some of what we make will change for years, and some of it will go quiet, and staying quiet will be the whole point.
A platform wants to be the ground you stand on, so that leaving means leaving everything. The lock-in is not a side effect, it is the business model wearing a friendlier word. We would rather grow a forest. Each tool stands on its own and does one thing you actually needed. They share a look and a set of manners the way trees in one wood share a climate, but none of them needs the others to live, and none is a trunk the rest depend on.
And a tool is defined as much by what we kept out as by what we let in. No is where we start. Every feature argues its way in against a standing objection and most lose, because most features are somebody’s good idea that would quietly make everyone else’s tool a little worse. Adding is easy and cheap to promise; removing later is expensive and rarely done. You cannot see the hundred doors we did not add. You can only feel that the room is not cluttered.
What it costs us
You cannot hand somebody a calm you do not have. A tool carries the weather of the shop that made it, and hurry is contagious. So the calm is not only a design choice, it is a way we have to live in order to keep it: forty-hour weeks rather than heroics, deadlines we set and can actually meet, sleep over one more push. We are not being precious. We simply cannot ship an unhurried tool from a shop that is on fire.
What crosses, and what does not
Lend somebody a book and they get the book. Not your other books, not the receipts, not the margins of everything else on the shelf. The loan has edges and nobody had to negotiate them. Software forgot how to do this: share a document and you share every revision, every comment you deleted, every version where the chapter was worse.
There are two ways to keep a note private from the person reading your manuscript. Send them everything and ask the software to hide part of it, or do not send that part. The first is a permission: ours to honour, ours to revoke, dependent on our continued good behaviour and our continued existence. The second is arithmetic. The note is not there. Nothing has to be trusted because nothing was transmitted, and we would rather build the version that needs less faith in us. A promise we make is a thing we could break, and the best way to be unable to break it is to have no way to.
That has a price, and we would rather say it here than have you find it in a support thread. You cannot un-send. What left, left. We can make the edges easy to draw and impossible to draw by accident; we cannot make them retroactive, and a revoke button would be selling the comfortable lie.
Most work worth doing has two people in it, and eventually two people stop agreeing. The usual answer is an owner and a set of guests, and guests can be removed, so a falling-out becomes an eviction and whoever lost the argument loses the work as well. Our answer is plainer and harder to live with: you both keep it. All of it, not half each. We do not arbitrate, because deciding whose it really was would make us a small court with no standing, ruling on somebody’s worst month with none of the facts.
People leave platforms over the neighbours, more reliably than over price or features or data. We decline to be the ground everyone stands on, so there is no lobby, no crowd we assembled, nobody you have been filed beside. A tool of ours holds your work and the people you named, and that is the whole guest list. The cost is the obvious one: we have no crowd to lend you. Nothing we make will find you an audience.
The machine, and the machine you own
We use these models and we are not going to sell them to you. Somebody who wants help with a manuscript wants their repetitions found and their commas fixed. “AI-powered” names the machinery and asks you to be impressed by it; proofread says what you get. So we name the work and never the mechanism. That is not concealment: the documentation says plainly what runs and where. The badge on the other side is the same move in a different coat, and we will not wear that one either. There is one line the tools do not cross, and the test is whether what the machine made leaves the room. A model reading your draft back so you can hear what your eye skipped is a tool. A model reading your book aloud to strangers is a performance, and a performance is a job somebody does.
Finally, the hardware, because a rule about respect that ignores the machine on the desk is a rule about nothing. We build for the computer you already have. That decides things a mission statement cannot: what a feature is allowed to cost, what we test on, how a page behaves on a slow connection. The machines here are second-hand for the same reason. Not thrift, calibration. If our machines are quick we stop noticing what the software costs, and the person at the far end pays for that comfort in seconds they will never get back.
With two limits, because a rule like that curdles into cheapness worn as virtue, and fastest when the machine is somebody else’s. There is a floor, and nobody should ever have to argue their way to a working computer. And repairable beats used: a second-hand machine nobody can open is worse than a new one with a service manual.
The oldest thing in the room should still be able to do the work.