the friction was never the music

It rained in Santa Monica this morning, so I put a microphone out the window.

That's one channel. The other is a MiniFreak, and both go straight into an H1n sitting on the sill. No computer, no session, nothing to undo. Thirty-nine minutes, improvised, one take.

That part is normal. That part I've done hundreds of times.

What's not normal is that by lunchtime it was on SoundCloud, on YouTube twice, and scheduled to Instagram — and I hadn't opened a single upload form. The last thing I posted before today was July 23rd. Six weeks of nothing, and it wasn't because I stopped making things.

where the energy actually goes

I want to spend my energy on music. That's the whole thing, and it's not complicated.

Distribution comes out of the same pot. It's the least interesting withdrawal I make, and it's the one that empties fastest — which is why I have a folder of finished work I never shipped. Ten song edits. Twenty-four reels. A film cut I've never shown anyone.

The gap was the twenty minutes between finishing something and anyone hearing it. Export. Resize. Write the caption twice because the platforms want different lengths. Upload the big file and watch a bar. Do it again for the next platform. Remember to write down where it went so you can tell later whether it did anything.

Twenty minutes isn't much. But it sits at exactly the wrong moment — right when the thing is done and the part of me that made it wants to go make another one. So I'd leave it for tomorrow, and tomorrow the next thing existed, and the old thing quietly became old.

I'm not beating myself up about that. It's just what happens when the boring part costs more than the interesting part earns.

YouTube I'd given up on outright. I took it off my calendar in September because posting there doubled the work for an audience I don't have yet. That was a real decision based on a real cost. It just wasn't a decision about music.

so I spent the day on the machinery instead

Not on a track. On the boring middle.

By the afternoon there was a path from a file on my desk to a live post: audio streams straight from storage to SoundCloud without ever sitting in a server's memory, video posts get scheduled through one queue, and — this is the part that actually matters — when something publishes, the row that planned it gets updated with the real URL automatically.

That last one sounds like nothing. It's the whole thing. My own notes had a standing rule to write down where each post went, and in the entire life of the project it had never once been done. Which meant six weeks of silence looked identical to six weeks of activity nobody recorded. You cannot fix a problem you have no evidence of.

it did not go smoothly, and that's the useful part

I want to be precise about this, because "I automated my publishing with AI" is the kind of sentence that sounds frictionless and wasn't.

The scheduler had shipped and was completely unreachable — it refused any post without a media file, and nothing in the app could attach one. It had never successfully sent anything. Three separate API routes were silently discarding fields: you'd save, get no error, and the data simply wouldn't be there. A counter that measures whether an upload got truncated reported zero bytes on an upload that had worked perfectly, because of a subtle mistake about how JavaScript copies objects — meaning a real truncation would have looked exactly the same. And the very first dispatch created three posts correctly and then reported all three as failures, because the response came back with a number where the code expected text.

That last one was the dangerous one. The posts existed. The system believed they didn't. One retry and I'd have posted everything twice.

None of that got caught by code being written quickly. It got caught by checking — reading the row back after the upload said success, comparing byte counts against the original file, opening the calendar to see what was actually there before hitting retry.

where I'm drawing the line

I'll say plainly what a machine did and didn't do today, because I think the distinction is the entire ethics of it.

It did not make the music. It didn't put the microphone out the window, or choose the morning, or play the MiniFreak. It didn't decide the piece goes quiet at twelve and comes back — that happened in a room, in real time, with rain, and I couldn't tell you why I did it.

What it did was find that a release of mine had been live on my own site for months with six tracks and no way to click and hear any of them. That two of my catalog links pointed at the wrong sets. That the date on my field recorder is a day slow, so everything it writes is misfiled. That two mixes had quietly swapped identities in my database, and the page had been showing the wrong length for one of them for who knows how long.

And that the clock on the H1n is a day slow — so the recorder that made this piece is the same device that had already misfiled it before I'd finished carrying it back to the desk. I'd never have caught that. I'd have just kept writing the wrong date on everything, forever.

That's not creativity. That's the accumulated unglamorous debt of a person who would rather be making things than filing them.

And the artwork on this one is a spectrogram — the actual frequency map of the recording, rendered in the same gold the site uses. I like that more than anything I'd have designed, because it isn't a decoration of the music. It's a picture of the music, the literal shape of thirty-nine minutes.

what I'd say to someone thinking about this

Automate the distance between finishing and sharing. Don't automate the finishing.

The moment this becomes something that writes the music, I've removed the only part I actually wanted to be doing. But the twenty minutes of export-resize-caption-upload was never mine in any meaningful sense. It was a tax on the thing I care about.

And it turns out I don't have a problem posting. I have a problem with what posting used to cost. When it's easy, I just do it.

It rained this morning. I put a microphone out the window. Then, for the first time, I actually put the result somewhere the same day.

That's the whole improvement, and it's bigger than it sounds.