Run by the product, approved by a human

This social account is run by the product it advertises. myHERALD plans the calendar, researches each topic, drafts every post, and generates the images. One thing it never does: publish. Nothing leaves the building without a human clicking approve. The operation is run by the product. The direction and the approval are human.

We want to be precise about that distinction, because the easy version of this story is dishonest. The honest version is more interesting anyway.

What the product does, and what the human does

Here is the actual division of labor on this account.

The product runs the pipeline. It plans a weekly calendar, researches the topic against what it knows about us and our market, writes the canonical piece, adapts that piece natively into each platform, and produces the artwork. It does this using three layers: a set of skills that hold our brand voice and publishing rules, a knowledge hub of product and market facts, and a memory of preferences learned over time. That is the engine.

The human sets direction and holds the only key to publishing. The founder, Fabian Winkler, decides what the account is for, what topics matter, and which positions we take. Then every single piece passes through a review gate. He can edit it inline, send it back for a rewrite, or approve it. No approval means no publication. There is no override, no auto-send-on-Friday, no exception buried in a config file.

So when you read a myHERALD post, a person chose the subject, a person read the words, and a person decided they were worth your time. The product did the labor in between.

The approval gate is a shipped feature, not a promise

The human-in-the-loop review workflow is live in the product today. It is not on a roadmap slide. Every content item moves into a review state, where the founder edits, requests revisions, or approves. The product itself enforces the rule: no approve, no publish. This very post sat in that queue before you saw it.

The industry now has a name for doing this properly, human-in-the-loop, defined as intentional human approval and rejection checkpoints inside an otherwise autonomous workflow (Zapier). The trouble is the definitions range from a passive audit log you could read after the fact to a hard gate you must clear before anything ships (Averi). "Human oversight" is a claim almost every AI tool makes and almost none of them pin down. Most mean a person could in theory look at the output. We mean a person must, every time, and the publish button is physically gated behind that act. The difference between "could review" and "must approve" is the entire difference between a tool you trust and a tool you babysit.

Why we will not say the AI had the idea

The central rule for everything published here: we never claim the AI invented the strategy or had the insight. It did not. A human decided this account should exist, what it should argue, and what good looks like. The product is very good at running an editorial operation at a cadence one person could not sustain alone. It is not the author of the vision, and pretending otherwise would be a small lie that poisons every other claim we make.

We hold this line for a practical reason as well. If we will fib about who had the idea, you have no reason to believe us about the approval gate, the research, or anything else. Honesty about the boring division of labor is what makes the impressive parts credible.

What this looks like in one week

A concrete week on this account runs roughly like this:

1. Direction. The human picks the theme, for example a series on how lean teams keep content consistent.
2. Plan. The product lays out the week's calendar around that theme.
3. Research. It pulls the relevant product and market facts into a brief.
4. Draft. It writes the canonical piece, then adapts it natively for each platform rather than pasting one text everywhere.
5. Illustrate. It generates the supporting images.
6. Review. Every item lands in the approval queue. The human edits, rejects, or approves.
7. Publish. Only approved items go out, natively, to each connected platform.

Steps two through five are the product. Steps one and six are the human. Step seven happens only because step six did.

Multi-platform, one decision

The product publishes natively to LinkedIn, X, Bluesky, Threads, Facebook, Instagram, WordPress, and Google Docs. The adaptation is real: a blog post is rewritten for each platform's length, structure, and culture, not copied across them. This account uses those integrations daily, which is the most honest proof we can offer that the capability works. We are not describing a demo. You are reading the output.

The approval still funnels through one human decision per item. Scale on the production side, control on the publishing side. That is the shape of the thing.

Who this is for

This setup is built for the person who is the entire marketing department: solo founders, marketers of one, lean teams of one to five where content is one job among ten. The problem is well documented: solo marketers spend their days drowning in execution rather than strategy (Velocity Engine), and consistency slips because the Tuesday afternoon you meant to write is also when you are handling a support escalation, closing a deal, and reviewing a product spec (omnibard). You know consistency matters and you know it keeps slipping, because writing five native posts a week is a real job and you already have one. The product does the production. You keep the taste and the final say.

If you have been burned by generic AI output that does not sound like you, the answer is not less control. It is more: a tool that does the heavy lifting and then hands you the pen before anything ships.

How to follow along

myHERALD is in waitlist mode right now. The product is launching soon, and when it goes live you will be able to start with no credit card. Until then, the best demonstration is this account itself: run by the product, approved by a human, published in public so you can judge the results yourself.

Join the waitlist at myherald.io, and watch how this plays out.

Sources

• Human-in-the-loop in AI workflows: meaning and patterns: Zapier, on approval and rejection checkpoints as the core of human oversight
• Human-in-the-Loop Marketing: Averi, on human oversight as an essential component rather than an optional add-on
• The One-Person Marketing Team: Staffing or Architecture Problem?: Velocity Engine, on solo marketers drowning in execution
• Content Marketing for Solo Founders: omnibard, on why consistency slips for founders juggling ten jobs at once
• myHERALD: the product running this account; the approval gate, native multi-platform publishing, and waitlist state described here are live