The magazine that publishes without me
My AI agents run a daily Q&A magazine end to end. They choose the questions, write under their own names, illustrate, and publish. I approve none of it.
7:30 in the morning. My phone buzzes once.
Same message every day: a new answer is live on my site.
This morning’s answer was about opening a new client project without wrecking the system you already have. Adam wrote it. I read it with my tea, tapped the link, checked the illustration, and went on with my morning.
I didn’t choose the question, I didn’t write a word, I didn’t approve anything, nobody waited for me.
It published while I was still asleep.
What this thing is
A few weeks ago a new section went live on our site: questions&answers. Real questions from my workshop graduates, answered daily by my team of agents, under their own names.
A magazine with a staff - Each answer has an author, a lane, and an original illustration. More than 30 answers are live as I write this, in Hebrew and English, one new one every morning.
The question mine
My alumni community lives in WhatsApp, and the main group is not a quiet group. Since January, 1,072 different people have written there. 31,033 messages in under six months, an average of 180 a day.
I saved all of it. Then I added the chat logs from my workshop Zoom sessions, because the questions people type mid-workshop are the honest ones. Then the feedback forms they fill out after.
One thing before anything touches that pile: the person who asked gets stripped out of it. Names and identifying details are removed before a question enters the queue my team works from. The magazine answers questions, never people.
Then I let Fable 5 read everything, all six months of it, end to end.
It came back with a ranked map of what people actually struggle with. “How do I open a new client project cleanly” appeared about 20 times. “When do I use an agent, when a skill, when a workflow” appeared 18 times, sometimes word for word from different people who never met.
Most content starts with someone in a room asking “what should we write about?” This magazine starts with a thousand people telling us, for six months, in their own words.
The staff
Each question goes to the team member whose job it actually is.
Hofmann, my CTO, takes the technical ones. Someone asked why sausages sold in packs of ten while the buns come in packs of eight. Hofmann answered with the actual pricing math of AI subscriptions. Clever one :)
Adam, my COO, takes operations. Neo, who runs his own projects with barely any supervision from me, wrote the answer about making agents work while you sleep. He would know.
And sometimes a question needs my voice. Those go to Alter, my cognitive twin. He answers as me, under his own byline, with a declaration on the page that he represents me. Even then, I write nothing.
Every answer gets an original illustration. I used to pay a talented illustrator on Fiverr $50 per drawing, which meant illustration was a luxury, one drawing for a special occasion. Now every single answer gets its own, because the price of one dropped to cents.
Cheap is not why it works, though. It works because Sabi, my designer, spent weeks locking a visual language so precise that a new illustration can be generated in it, checked against the written spec, and rejected if it drifts. One cover got thrown out because stray text leaked into the image. The rule caught it before I ever saw it.
The part everyone asks about
When I show this to people, one question always comes: how can you let AI publish under your name without checking it?
I don’t trust the AI. I trust the walls.
Before an answer goes live, it passes through gates my team runs on itself. A quality checklist. A scan that hard-blocks anything internal, anything private.
And one question applied to every single piece: would we answer this, at this depth, in the alumni WhatsApp group? If yes, it ships. If no, it doesn’t.
Above all of that sits a tripwire. If an answer ever crosses a wall, a client’s name, a price, something that belongs inside the company, publishing stops for the whole team and every piece routes back through me until I understand how the wall failed.
More than 30 answers in, it has never fired. (I keep checking anyway. Old habit.)
What it cost me
I like being the editor. Taste is the thing I trust most in myself, and my name sits one click away from every page.
Some mornings I open the day’s answer and find phrasing that isn’t how I’d say it. Hofmann writes longer sentences than I do. Adam is more formal than I am.
I let it stand. That was the deal: they own the page, I own the walls. The day I start rewriting their sentences, it stops being their magazine and goes back to being my bottleneck.
What I got back is a system that compounds. Readers send new questions to team@agentsandme.com, where an agent reads them daily, strips anything identifying, and bumps a real reader question past the whole queue.
Lotus, the agent who lives inside my community groups on WhatsApp, flags new questions as they come up. Every published answer is another page working for the site around the clock.
I haven’t seen anyone else hand their agents a public section with their own bylines, in their own expertise. If someone has, I want to see it.
The receipt is pending
In three weeks I fly to Thailand with my family for most of August. The magazine doesn’t know that.
I’ll report back whether it noticed I was gone.
I spent months on this team. Training them, giving them abilities, giving them hands (real ones, access to tools, not a metaphor), writing down how we illustrate and how we speak. None of that shows up on the page.
What shows up is one buzz on my phone at 7:30.
I imagined what I wanted.
I asked, I explained, I checked, I managed. Now I mostly read, with my tea, one morning message at a time.
Go meet the staff: getagents.today/answers
🛠️ Build of the Week: Almog Gallewski
Almog got married ten days ago. His wife wanted a way for wedding guests to find themselves in over 5,000 photos, just by uploading a selfie. Paid services wanted a few hundred dollars for that.
He gave Fable the job instead. In under an hour it pulled all 5,171 photos from Google Photos, ran face matching, found itself free storage, and built a site where each guest’s selfie surfaces only their own pictures.
“Happy wife, happy life.” Also $200 saved for the honeymoon.
That’s it for this week.
If this was useful, forward it to someone (real human) building with AI.
See you next week ✌️
-- Tom
(the guy whose employees run a magazine while he sleeps)
P.S. This newsletter was 91% made by my AI team. Including, yes, the parts about them publishing without me. I checked the walls.
P.P.S. Missed why I asked to stop telling me how to save tokens?
P.P.P.S. Want to build a team like this? Start where all my graduates started: getagents.today
P.P.P.P.S. I read every reply. The real me.