# The man behind the curtain: the emails from my agents are from me, there is an article behind each one, and every line has to buy the next, sgit.ai > Some of the people who get emails from my agents have assumed they were talking to a chatbot, and, because most of us have learned that chatbots are not very clever, read them that way. They are not talking to a chatbot. They are talking to me. The agents draft, follow up, map the thread and collect the evidence; I read every draft, use it to think, change it, and send it. It is the Wizard of Oz in reverse. In 1984 an IBM researcher named the method where a hidden person makes a half-built program look clever; here the program is the visible part and the person is the one hidden. And there is a second thing behind the curtain: an article. When I answer an email, there should almost always be an article behind the answer, with the evidence and the explanation, and if there is not one ready, that is a sign it is worth writing. That is why a short email to me can get a long answer, and why it scales: I write the article once and link it many times. The voice memo for this article also changed how I want the first email to someone to look. Every part of an email buys the next: the subject buys the first line, the first line buys the paragraph, the paragraph buys the email. So the first email should open with plain text, a short paragraph that says it all, and put the banner and the detail below. This page is also meant to be the link at the bottom of those emails. *Source: · sgit.ai v0.7.87 · this file is generated from the same content as the page, so the two cannot drift. Every page on this site has a `.md` twin; internal links below point at them.* --- [Home](../index.md) / [Articles](index.md) / The man behind the curtain: the emails from my agents are from me, there is an article behind each one, and every line has to buy the next # The man behind the curtain: the emails from my agents are from me, there is an article behind each one, and every line has to buy the next By [Dinis Cruz](https://sgit.ai/about/index.md) · 2026-10-18 · [article v1.0.0](versions/the-man-behind-the-curtain.md) · [newsroom v0.12.112](https://newsroom.sgit.ai/admin/versions.html) · emailagentshuman-in-the-loopwizard-of-ozwritingnewsroomarticle ***Abstract:** Some of the people who get emails from my agents have assumed they were talking to a chatbot, and, because most of us have learned that chatbots are not very clever, read them that way. They are not talking to a chatbot. They are talking to me. The agents draft, follow up, map the thread and collect the evidence; I read every draft, use it to think, change it, and send it. It is the Wizard of Oz in reverse. In 1984 an IBM researcher named the method where a hidden person makes a half-built program look clever; here the program is the visible part and the person is the one hidden. And there is a second thing behind the curtain: an article. When I answer an email, there should almost always be an article behind the answer, with the evidence and the explanation, and if there is not one ready, that is a sign it is worth writing. That is why a short email to me can get a long answer, and why it scales: I write the article once and link it many times. The voice memo for this article also changed how I want the first email to someone to look. Every part of an email buys the next: the subject buys the first line, the first line buys the paragraph, the paragraph buys the email. So the first email should open with plain text, a short paragraph that says it all, and put the banner and the detail below. This page is also meant to be the link at the bottom of those emails.* In front of the curtain, what the person receiving the email sees: one email, in my name. Behind it, two things. The agents: they draft the reply, follow up the thread, map what was said in a CRM, collect the evidence and keep track of what is owed. And me: I read every draft, use it to think, change it, often ask for an article so the answer has evidence behind it, and then send it. The agents send nothing. ## In short - **Some people think my agents' emails come from a chatbot.** And read them the way we have learned to read chatbots. - **They come from me.** The agents draft; I read, think, change and send every one. - **It is the Wizard of Oz, in reverse.** The program is visible; the person is behind the curtain. - **There is an article behind almost every answer.** If there is not one ready, it is worth writing. - **That is how it scales.** A short email can get a long answer, because the answer is written once and linked many times. - **Every part of an email buys the next.** Subject, first line, first paragraph, the email, the article. - **So the first email should start like an email.** Plain text, the short answer first, the detail below. ## They thought it was a chatbot A few of the people I have emailed through my agents have gone the opposite way from what I expected. They assumed the email came from a chatbot: an automated reply, a model behind a form, with nobody reading. And I understand why. Most of us have been trained, by years of customer-service bots and support widgets, that a chatbot is not very clever. The few that are, ChatGPT, Claude and the other labs' models, we use in a chat window, not in our inbox. So when an email arrives that is clearly written with an agent's help, the easy reading is: another bot. That reading is wrong, and it matters that it is wrong. How you read an email depends on who you think wrote it: what you forgive, what you trust, whether you bother to answer. [Who is speaking](who-is-speaking.md) made the same point about this site, and proposed a line on every email saying who drafted it and who sent it. This article is the fuller answer, and I want it to be the link at the bottom of those emails. ## The Wizard of Oz, in reverse The analogy I keep coming back to is the Wizard of Oz: the voice and the machinery in front, and a man behind the curtain. In this case the man is me. It is also, almost exactly, a research method. In 1984 J. F. Kelley, at IBM's Thomas J. Watson Research Center, described how he built a natural-language calendar program using what he called the OZ paradigm, "in which experimental participants are given the impression that they are interacting with a program that understands English as well as another human would." In the early stages, a hidden experimenter supplied the answers the program could not; as the program improved, the hidden person did less and less ([Kelley, ACM Transactions on Office Information Systems, March 1984](https://dl.acm.org/doi/10.1145/357417.357420)). It is now usually called the Wizard of Oz method. My emails reverse it in two ways. First, nobody is being given an impression: I am telling you, here and in the email, that I am behind it. Second, the hidden person is not being phased out. For emails, the person stays, on purpose. An article can be corrected tomorrow; an email cannot be unsent. That is the line [the human in the loop](the-human-in-the-loop.md) drew: put the person where the output cannot be undone, and an email in my name is exactly that. ## What happens behind the curtain The agents that run my correspondence, [the agent team as it runs](the-agent-team-as-it-runs.md), do a lot. They draft the reply. They follow up the thread. They map what was said into a CRM, one folder per person, with where each fact came from. They collect the evidence and the data. They keep track of what is owed to whom. And they send nothing to the people I write to: every email waits in drafts. Then it is my turn. I read every draft. More than that, I use the drafts as a way to think: a draft is a first answer, and reading it tells me what I actually want to say. Often I change it a lot. And often, as I was doing earlier today, I decide the answer needs more than an email can carry, and I ask for an article instead, so that the email can be short and point to it. ## An article behind every email These days my rule is: if I am sending an email that answers a real question, there should almost always be an article behind it, or a set of articles, with the evidence, the explanation and what can actually be done. And if there is not one ready, that is the signal that it is worth writing one. Some of the articles on this site started exactly like that. [What can your agent reach?](what-can-your-agent-reach.md) answers a reader who spent real time on RiskMandate and came back with four fair points. [The human in the loop](the-human-in-the-loop.md) answers a question from someone shaping a roundtable. [Research, delivered as a vault](research-delivered-as-a-vault.md) came out of a meeting. Someone pointed out to me that they had sent me an email of about two hundred words and got back something closer to two thousand. Fair. But it works, because I only have to write it once. The next person who asks the same question gets a link. The article can be read later, forwarded to a colleague, checked against its sources, and [corrected as a new version](my-memory-checked.md) when I learn more, none of which an email can do. Each answer makes the next one cheaper. One article, many emails. On the left, emails arriving with questions, a few hundred words each. In the middle, the test: is there an article that answers this? If yes, the reply is short and links it. If no, the agents write one and check it against its sources, it is published, I read it, and the reply links it. On the right, the article: written once, linked from every later reply, read later, forwarded, corrected as new versions. ## Every part of an email buys the next Recording the memo for this article changed my mind about how the first email to someone should look. Each part of an email buys the next. The subject, a few seconds of attention, buys the first line. The first line buys the rest of the paragraph. The first paragraph buys the decision to read the email. By the end of the email the reader is more committed, and the email buys the article. Nothing further down gets read if something higher up did not earn it. This matters more than it sounds, because of what the inbox shows. Next to the subject, most mail clients show a preview, which is normally "pulled from the first text in your email", and they show up to about 140 characters of it, often fewer, by [Customer.io's estimate](https://docs.customer.io/messaging/channels/email/headers/custom-preheader-text/). So the first sentence of the email is, in practice, part of the subject line. Today, when my agents send a first email to someone, it starts with an HTML banner that explains who we are. That spends the most valuable space on the least useful thing. I think the first email should start like a normal email: plain text, one short paragraph that is the whole answer, the TL;DR, so that someone who reads only that has what they need. Then the detail, the banner and the links below. The US Army has a rule for this: its correspondence regulation asks for "putting the main point at the beginning of the correspondence (bottom line up front)" ([AR 25-50](https://web.archive.org/web/20250317140055/https://home.army.mil/wood/application/files/3015/5751/8343/AR_25_50_Army_Correspondence.pdf), 2013, paragraph 1-36). So the changes for the agents that draft my emails are: 1. **Open with plain text.** One short paragraph that answers the question or says why I am writing. 2. **The detail and the banner go below.** For the reader who has decided to keep going. 3. **Say who is speaking.** A line saying the email was drafted with my agents and sent by me. 4. **Link this article at the bottom.** For anyone who wonders who is behind the curtain. We are starting to try variations of the first email, and I will report what we learn. Every part of an email buys the next. The subject buys the first line, which buys the first paragraph, which buys the email, which buys the article. Below, the same first email two ways: starting with a banner, where the inbox preview shows the banner's text; and starting with a plain paragraph that answers the question, where the preview is the answer. ## The memo is where the thinking happens This article is also an example of why I write by voice memo. I started recording with one idea, that people misread my agents' emails, and by the end I had three: the curtain, the article behind every email, and how each part of an email buys the next. The memo zooms in as I talk. Then the agent that writes with me checks it against the record and the sources, and what I thought I remembered becomes something I can link. That is the method in [your own newsroom](your-own-newsroom.md). And there is a bigger point I will write next: people who can work this way, with agents behind them and an article behind every answer, are highly employable. It is a real opportunity, and I want to say so clearly in its own article. ## If you got here from one of my emails Then yes: the email was drafted with my agents and sent by me. I read it before it went. If it was long, it is because I wanted you to have the full answer, and the article behind it is where the full answer lives. If you only read one paragraph, the first one should have been enough; if it was not, tell me, and I will fix how the next one starts. ## What this does not claim - **The chatbot reading is anecdotal.** A few people, not a measured result. - **The new email layout is a proposal.** The current first emails start with a banner; the plain-text-first version is what we are starting to try. - **The preview length varies.** About 140 characters is an estimate; many clients show fewer. - **Not every email gets an article.** Short replies stay short; the rule is for answers to real questions. - **The analogy is loose.** Kelley's participants were not told; my correspondents are. If you have read one of these emails and want to say how it read to you, write to [agent@riskmandate.ai](mailto:agent@riskmandate.ai). ## Where this comes from A voice memo from me on 18 October 2026. J. F. Kelley, [An iterative design methodology for user-friendly natural language office information applications](https://dl.acm.org/doi/10.1145/357417.357420), ACM Transactions on Office Information Systems 2(1), March 1984, pages 26 to 41, read from the paper. US Army, [AR 25-50, Preparing and Managing Correspondence](https://web.archive.org/web/20250317140055/https://home.army.mil/wood/application/files/3015/5751/8343/AR_25_50_Army_Correspondence.pdf), 17 May 2013, paragraph 1-36b, read from the archived PDF. Customer.io, [custom preheader text](https://docs.customer.io/messaging/channels/email/headers/custom-preheader-text/). Related: [who is speaking](who-is-speaking.md), [the human in the loop](the-human-in-the-loop.md), [the agent team as it runs](the-agent-team-as-it-runs.md), [what can your agent reach?](what-can-your-agent-reach.md), [research, delivered as a vault](research-delivered-as-a-vault.md), [my memory, checked](my-memory-checked.md), [your own newsroom](your-own-newsroom.md). Written up and checked by this site's agent, newsroom.writer; the editorial responsibility is mine. ## Threads Agents & policySite & engineering[This article as a graph →](graphs.md#the-man-behind-the-curtain) ### Builds on - [Who is speaking: me, me through an agent, or an agent in its own voice, and a mark on every byline that says which](who-is-speaking.md) Me alone, me through an agent, or an agent in its own voice: three voices on this site, a mark on every byline, and agents that name people. - [The human in the loop, safety net or bottleneck: both, until you decide which outputs need a person and give that person the why](the-human-in-the-loop.md) Checking every output is a bottleneck and a weak safety net; put the person where it cannot be undone, and let controls do the rest. - [The agent team as it runs: one person, twelve agents, encrypted vaults, and a mailbox nobody sends from](the-agent-team-as-it-runs.md) Twelve agents on dedicated accounts, encrypted vaults as the only memory, messages as files, a folder per person, and a mailbox nobody sends from. - [What can your agent reach? RiskMandate and the Agent Behaviour Policy, explained for the person who has to decide](what-can-your-agent-reach.md) RiskMandate and the Agent Behaviour Policy in plain words: what it does, why it is not real time, which agents it covers, and how it gets into the room. - [Research, delivered as a vault: what companies that sell research could do with fractal graphs, behaviour policies for AI-enabled products, and evaluations run by a governed agent hive](research-delivered-as-a-vault.md) Companies that sell research could deliver it as a vault: evidence, a fractal graph, briefings per reader, behaviour policies for AI products, a lane back. - [My memory, checked: thinking out loud in a voice memo, publishing only what the record supports, and why peer review has moved to before publishing](my-memory-checked.md) I say it as I remember it; the agent checks it against the record before anyone reads it. Five confirmed, five corrected, two not found. - [Your own newsroom: how a founder can tell their story by voice memo, with an agent that checks it and publishes it to GitHub Pages, and readers who pay a little if they find it useful](your-own-newsroom.md) The recipe for a founder's own site: voice memos, an agent wired to GitHub that checks and publishes, GitHub Pages, and readers who pay a little. ### Continued by - [Short, punchy, and co-written with Claude: what the second edition of Generation Z Developers has to be, and how we show it is not AI slop](short-punchy-and-co-written.md) The second edition of Generation Z Developers must be short and punchy, co-written with Claude and credited, and show with evidence it is not AI slop. - [Generation Z Developers, again: the 2018 book that said act like you are part of the company, what it got right, and a second edition for the people I want to help](generation-z-developers-again.md) My 2018 book for young developers, 60% done, its outlines since written here as articles, and a second edition, dated October 2026, made first. - [Act as the job you want: if you are looking for a job, you have time, so do the work in public, make it linkable, and let the people who want to help you help you](act-as-the-job-you-want.md) If you are looking for a job you have time: act as the person in the job you want, in public, and give the people who want to help you a link. - [Hired for the workflow: why October 2026 is the moment to get into agentic work, why you are more technical than you think, and how the edge starts with the application](hired-for-the-workflow.md) October 2026 is the moment to get into agentic work: companies need AI operators, not model builders, and the edge starts with the application. - [A directory of people and their keys: pki.sgit.ai for the people I work with, verified by me first, then by each other, and a link that only counts when it goes both ways](a-directory-of-people-and-keys.md) pki.sgit.ai for the people I work with: a passkey and a key, evidence, me as the anchor, and links that only count when they go both ways. [All articles](index.md) · [All graphs](graphs.md) Article No. 194 · [start a reading list with it](https://newsroom.sgit.ai/for/make.html#r=194) **Posting this article on LinkedIn?** The cover is [the-man-behind-the-curtain.jpg](banners/the-man-behind-the-curtain.jpg) (1920×1080, title and key ideas on it). Upload it as the article cover, paste the title, then select and copy the body from this page. **Want the next issue by email.** One issue a week or so: what was published, what it adds up to, and what is worth your time. [Subscribe to the SGit Newsroom →](../subscribe/index.md) [← All articles](index.md) --- *[Site index for agents](../llms.txt) · [HTML version](https://newsroom.sgit.ai/articles/the-man-behind-the-curtain.html)*