Teardown No. 130 · Community

31,033 WhatsApp Messages: How Tom Even's agents turned his workshop alumni's questions into a magazine he doesn't approve

Tom Even teaches workshops on running a team of AI agents and runs his own small company with 40+ of them. He took the 31,033 messages his alumni WhatsApp group wrote in under six months, plus Zoom chat logs and feedback forms, stripped out names and identifying details, and had a model rank what people actually struggle with. His agents now answer those questions in a public questions&answers section on his site, under their own bylines, with an original illustration each. He does not write or approve the answers. He relies on a quality checklist, a scan that blocks anything internal or private, and a tripwire that stops all publishing if an answer crosses a line.

Try this step

Mine the community you already have. He saved the whole WhatsApp history, added the chat logs from his workshop Zoom sessions, "because the questions people type mid-workshop are the honest ones", and added the feedback forms people fill in afterwards.

The problem

His alumni group “is not a quiet group.” Since January, 1,072 different people had written there, an average of 180 messages a day. The same questions kept coming back. In a later monthly review he gave himself a rule for this: “Been asked the same question more than three times?” Turn it into something permanent. The catch is his time. Every answer he writes or edits himself puts him back in the queue.

The blueprint (as he describes it)

  1. Mine the community you already have. He saved the whole WhatsApp history, added the chat logs from his workshop Zoom sessions, “because the questions people type mid-workshop are the honest ones”, and added the feedback forms people fill in afterwards.
  2. Strip the person before anything else. Names and identifying details are removed before a question enters the queue. In his words, “The magazine answers questions, never people.”
  3. Rank by what repeats. He let Fable 5 read all six months end to end. It came back with a ranked map. “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.
  4. Route each question to the agent whose job it is. Hofmann, his CTO agent, takes technical questions. Adam, his COO agent, takes operations. Questions that need his own voice go to Alter, his “cognitive twin”, who answers under his own byline with a declaration on the page that he represents Tom.
  5. Write the illustration style down. He used to pay an illustrator on Fiverr $50 per drawing. Sabi, his designer agent, locked a visual language so a new illustration can be checked against the written spec and rejected if it drifts. One cover was thrown out because stray text leaked into the image.
  6. Gate on walls, not on taste. Before an answer goes live it passes a quality checklist and a scan that hard-blocks anything internal or private. One test applies to every piece: would we answer this, at this depth, in the alumni WhatsApp group? Above that sits a tripwire. If an answer ever includes a client’s name, a price or anything internal, publishing stops for the whole team and every piece routes back through him.
  7. Keep the intake open. Readers can send new questions to a team address. An agent reads them daily, strips anything identifying and moves a real reader question ahead of the queue. Lotus, the agent inside his WhatsApp groups, flags new questions as they come up.

Stated results

  • At launch (Jul 14, 2026): “More than 30 answers are live as I write this, in Hebrew and English, one new one every morning.” The tripwire had “never fired.”
  • Live page (checked 2026-11-20): the English index calls itself “No. 76” and says the magazine is “written, designed, and published by the agents&me agent team, entirely on its own.” By the dates on its cards: 45 answers in July, 14 in August, 6 in September, 5 in October and 6 in November so far, the latest dated Nov 11, 2026.
  • His time: he reads one notification at 7:30 in the morning. Some phrasing is not how he would say it, and he lets it stand: “they own the page, I own the walls.”
  • Traffic, sign-ups or workshop sales from the magazine: None stated.

Caveats

  • The daily cadence did not hold on the English page. After July the dated entries drop to single digits per month. The Hebrew side was not counted.
  • He planned to test whether the magazine noticed him gone during a family trip in August. In his September review he wrote that the trip was cancelled because he was still sick.
  • The same review says workshops became harder to fill, and that about 40% of sign-ups came through big podcasts he was interviewed on. None of that is attributed to the magazine.
  • The Fable 5 model is named in his post. No other models are named for the answer agents.
  • He sells workshops (getagents.today) and the newsletter promotes them. He notes the newsletter itself is mostly made by his agent team.
  • Handles: LinkedIn in/tom-even-ai (from the newsletter’s about page). No verified personal X, Bluesky, Threads, Instagram or TikTok profile was found.

FAQ

Where do the magazine’s questions come from?

From his alumni WhatsApp group (31,033 messages from 1,072 people in under six months), his workshop Zoom chat logs and feedback forms, plus new questions sent in by readers. Identifying details are removed first.

Does Tom Even review the answers before they publish?

No. The live page says he does not write or approve them. A quality checklist, a scan for internal or private material and a tripwire that stops all publishing are the controls.

Is it still publishing every day?

Not on the English page. It listed 45 answers in July and 5 in October, and the latest one is dated Nov 11, 2026.

Has it brought in traffic or customers?

None stated. He gives no traffic, sign-up or sales numbers for the magazine.

Sources


Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.