The challenge

This one is ours. Built on the system we run ourselves: in production every week, and this site is its output.

We sell automation into the US and Northern Europe, which means publishing in English on the subjects our buyers care about — every week, not when someone gets around to it. We had the same problem our clients have, and we had watched it fail in both directions.

Content that depends on a free afternoon stops the first busy week. The plan holds while things are calm, then a customer emergency eats Friday and the calendar becomes a document nobody opens.

Content a model writes without supervision fails differently, and worse. It drifts off message, it invents numbers, it reads like everything else on the internet. Silence costs you nothing. Published nonsense costs you the buyer who was reading.

So we built the thing that solves both, pointed it at ourselves first, and ran it long enough to know what it costs to operate.

What we built

A weekly cycle that turns one research pass into a full channel pack, with a person in front of the publish button.

Backlog. Ideas land continuously, from scheduled research runs and from whoever has one. One entry point, one file, no parallel lists.
Research. Once a week the machine takes the strongest two or three ideas and works them into a dossier: sources, data worth citing, the angle, what has already been said.
Pillar. The dossier becomes one substantial piece — the thing that is actually worth reading.
Derivatives. The pillar becomes the pack for every channel: the blog article, the newsletter version, three LinkedIn posts, an X article, an X thread, a video script. Eight pieces from one research cycle. That is the written template, not a round number we liked.
Schedule. Every piece has a day and a channel before the week starts.

Three things the machine is not allowed to do, by design:

It does not decide the message. Positioning and the catalogue live in fixed source files. The machine writes from them and has no licence to reinterpret them.
It does not invent numbers. Every figure it may use is on a list, with where the figure came from. Anything not on that list does not get written.
It does not publish. A person reads the pack first — roughly one to two hours a week — and approves, cuts, or sends a piece back. The gate itself is a build of its own: Publishing With a Human Tap.

That gate is also where the system gets better. What gets cut, and the reason it was cut, goes back into the brief for the next cycle, so the following week starts closer to publishable. The human is not a rubber stamp on the output; they are the input for the next run.

Ours runs on a schedule in our own cloud account. For a client, the same engine runs in their own cloud account, on their channels, writing from their message source.

The results

The engine has been producing a full pack a week. The claim is narrow and we will keep it that way: what is contracted is throughput.

Number	What it measures
1 → 8	pieces produced from one research cycle
Weekly	one cycle, every week, on a schedule
~1-2 hrs	of human review per weekly pack

What this engine does not promise is rankings, traffic or leads, and we do not publish our own numbers on those against it. Content shipping on schedule is a precondition for those outcomes, not a substitute for them. Anyone who sells you the second as if it were the first is quoting you a forecast, not a system.

The measurable part is the simple part: the pieces exist, on time, every week, and a person read each one. If a blog post, a newsletter or a LinkedIn post brought you to this page, it came out of this cycle.

What you get

The engine on your channels: the backlog, the weekly research pass, the pillar, the derivative pack and the publication schedule — running in your own cloud account, writing from your message source, with the approval gate in front of it. Process audit first, scope in writing, and 30 days of support.

Publishing only when someone has a free afternoon? Start with the free AI Audit, or book a call.