I have a complicated relationship with X.
It is where the people I learn the most from think out loud: engineers shipping AI products, founders writing about what broke, researchers posting the paper a week before everyone else talks about it. It is also where my evenings went to die. Open the app to check one thing, look up, forty minutes gone, and the one thing still unchecked.
I did not want to quit it. I wanted the signal without the slot machine.
Willpower lost, every time
I tried the usual things first. App timers I dismissed. Deleting the app, then reinstalling it on the train. Turning my phone boring on purpose, which helped with everything except the one app I actually wanted to read.
The problem was never the content. It was the shape. A feed has no last page. It is built so that "one more" is always available, ranked by whatever keeps your thumb moving, and every refresh is a small lottery ticket. You cannot out-discipline a system that an entire company tunes every day to beat your discipline.
So I stopped trying to use it less, and changed the shape instead.
Newspapers ended
A newspaper has something no feed has: a back page. You read it, you fold it, you are done. Nobody refreshes a newspaper.
That became the whole design. Twice a day, everything worth my attention from X, collected, edited and printed into something finite. Delivered to a device that cannot do anything else.
I called it Pocket Post.
How it works
Twice a day, at 06:00 and 18:00, a small job runs. A morning edition and an evening edition, like papers used to have.
- Collect. It gathers everything since the last edition on the subjects I care about: AI, engineering, leadership, and the rest of my interests.
- Edit. Claude Opus acts as the editor. It groups the posts into sections, merges the ten takes on the same announcement into one story, writes a short summary for each, and keeps the link to every original post.
- Cut. It drops what a good editor would drop: engagement bait, reply wars, "this changes everything" with nothing behind it, and anything I have already seen.
- Print. It lays the result out as a small newspaper, with a front page, sections and a back page, and saves it as an EPUB. One section is just for fun: what is going on in whichever city I happen to be in that day. It changes when I travel, and it is the part I read first.
- Deliver. It emails the issue to my PocketBook e-reader, where the morning edition is waiting with my espresso and the evening edition with the end of the workday.
The rules I gave the editor matter more than the code:
- It has to end. One issue takes twenty to thirty minutes to read front to back. If a day was busy, the editor cuts harder; it does not add pages.
- Every story points back at its source. Summaries are for deciding what to read, not a replacement for reading it.
- No outrage section. If a post exists mainly to make me angry, it does not make the paper.
- Long reads get their own page. The threads and articles worth twenty minutes are listed at the back, for when I actually have twenty minutes.
What changed
The difference is not that I read less. Some days I read more, and better: the paper gives me the whole argument instead of the loudest reply to it.
The difference is that it ends. I finish the back page, the e-reader goes on the shelf, and there is nothing left to refresh. X became a place I post, not a place I live.
I do not count it as screen time, and I do not measure it like screen time. It is closer to me time: half an hour with zero notifications, on a screen that cannot send me one.
And because e-ink cannot autoplay a video or flash a notification at me, the reading is calmer too. It feels less like being fed and more like being informed.
What I took from it
Change the shape, not the willpower. Most bad habits I have broken, at home or at work, went the same way. Telling yourself, or your team, to try harder against a system designed to win rarely works. Redesigning the system does.
The model is the smallest part. The language model that edits Pocket Post is the least interesting piece of it. What makes it useful is everything around it: where the posts come from, the rules the editor follows, the format it prints to, the device it lands on. That is the same lesson I keep relearning with AI at work. The value is in the harness.
Build the tool you wish existed. The first version took one evening. It has been refined, a rule here and a section there, over months. It costs nothing extra: it runs on the Claude subscription I already pay for. A few years ago it would have been a product someone had to sell me. Now it is an evening project, made for exactly one reader.
If you have your own feed that eats your evenings, try giving it a back page.