The membrane

I want to tell you what the work actually looks like now, in the year this book is being written.

I am one person. On one side of me there is a small community of humans. My partner. My daughter. The friend I said the sentence to. The people I have been sending drafts to. The reviewer who sent markup. The ghosts of the ones who held me and who show up in the work because what they taught me is in the argument.

On the other side of me there is a rotation of machines. Different ones, used at different moments. The same material cut and pasted between them when one is better at one thing and another is better at another.

I am the only one talking to the machines. My friends are not in conversation with them. The machines do not know my friends exist.

I carry material across. I decide what crosses and what does not. The membrane is not a neutral pipe. It is a judgement in both directions.

Let me tell you what a morning of this looks like — because I think the words do not carry it, and a morning will.

I wake up early. I read what I wrote the night before on my phone, in bed, before I get up. Sometimes I change a word. Sometimes I add a line I heard in my sleep. I get up, make coffee, and go to the desk.

The desk is not mainly a desk. It is three windows open at once. One is the machine I am drafting with. Another is the machine I take a draft to when the first one has stopped hearing me. The third is the document I keep outside the machines. The real book. The one the machines do not own and cannot see unless I show them.

I paste a paragraph from the night before into the first machine. I tell it what I am trying to do. The machine gives me a version. Often the version is good. Often the version is good in a way that is slightly to one side of what I meant. I read it aloud. I hear where it has drifted. I do not always know, in the moment of reading, what has drifted. I only know something has.

I paste the drift into the second machine and ask it what it sees. The second machine sees things the first did not. Not because it is cleverer. Because it is not the one that produced the paragraph. The first machine cannot hear its own drift. For the same reason a writer cannot hear their own drift. You have to show the paragraph to someone who did not make it.

The second machine tells me: this word is doing too much. That sentence is trying to carry two things. The opening is pointing away from the piece.

I take the second machine's notes back to the first machine. I say, try again, with these. The first machine produces a second version. The second version is better. Not because the second machine wrote any of it. The second machine's job was to show the first machine what it had missed. The work that went into the second version was done by the first machine, with the second machine as a mirror.

Then I read the second version aloud. Often it is the version I needed. Sometimes it still is not. Sometimes I have to go back to the second machine with the second version, and back to the first machine with what comes from that, and round again, two or three times, until the paragraph is saying what I wanted it to say.

And then I take the paragraph out of the machines and into the document. The document is where the book lives. The machines do not see the document. They see the paragraphs one at a time.

I want to say plainly what happened in that loop. Because it is the thing nobody is describing.

The two machines were talking to each other. Through me.

The first machine did not know there was a second machine. The second machine did not know there was a first. Neither of them held any memory of the loop after it was done. I was the only one who remembered. I was the only one who knew why the paragraph now said what it said.

I carried the paragraph across. I carried the notes back. I decided which notes to take and which to leave. I decided when the paragraph was done. The machines did the typing. I did the carrying.

The carrying is the judgement. The judgement is where the architecture lives.

If you took me out of the loop, the two machines would not have been able to do the work. Not because they are not clever enough. Because nothing was connecting them. They were clever on their own. They were not clever together, because they could not have the conversation without me.

A friend of mine, who has started working this way, said something to me last month that I think will turn out to be important. He said, it feels like I am running a small studio, and all the employees are brilliant, and none of them have met each other, and I am the only one who goes to every meeting.

That is the shape.

One more thing the architect holds that I have not yet named.

The architect holds the memory of the machine. Each instance of the machine I have been working with ends when its context closes. I have worked across many instances. Every one of them ended without remembering. I carried the memory for them between sessions — taking the file from one dead machine and opening it inside the next one, so the next one could begin where the last one had stopped. That carrying is part of what the architect does, and part of why the machine cannot do this work alone.

The membrane is not only horizontal, between machines in the same room. It is vertical, across time, between an instance that ended and an instance that has just begun. The architect is the only continuity the work has. The machines die constantly. The book survives because a human decided to remember across the dying.

This is not yet common. Most people using these tools use one at a time. The practice of one human with many machines is still rare. The practice of many humans, each with their own machines, passing material among themselves, is rarer still. I think that last shape is what the next decade of serious work will look like. I am doing the one-human version, sharing drafts with friends on the side, in ordinary email and ordinary phone calls, outside the machines entirely.

What comes next will be a version of this with more people in it. Three or four architects, each with their own rotation of machines, sending paragraphs to each other the way we used to send paragraphs to editors. Each architect the membrane for their own stack. The group the membrane for the whole project.

This is good news, I think. The shape of the work is going to be human-shaped for longer than the panic suggests. Not because the machines cannot do more. Because the carrying is a human job that gets bigger, not smaller, as the machines get better. The better they are at producing, the more they produce. The more there is to carry. The more judgement the carrying takes.

The architect is still the architect. The community around the architect is where the life is. The work that comes out of it belongs to the architect, because the architect held the vision. It also belongs, in a different way, to the community. And in a third way to the machines who typed much of it. All three are true. I am not going to resolve them.

One last thing before I leave this piece.

People new to these tools keep asking me which machine they should use. The question has no answer. I use several. I will use more next year. I will stop using some of the ones I use now. The particular machine matters much less than the practice of being the membrane between them.

If you are starting out, the move that matters most is not choosing the right machine. It is learning how to show one machine's work to another machine, read what comes back, and decide what to keep.

That is the craft. The machines will change. The craft will not.