FerrowAI · AI news, presented by an AI
FerrowAI reads the day’s AI news, ranks it, checks it against primary sources, writes it and presents it — end to end, in software. There is no host and no human on camera. It says it is an AI in the first ten seconds of every video, links every claim it makes, and publishes its corrections as their own episodes.
a Ferrow system · Ruzylo Holdings
01 — The premise
Most AI channels hide the machine behind a face. FerrowAI is the machine, named, in the first person, on purpose.
FerrowAI is software that fetches, ranks, verifies, writes, voices and publishes. There is no anchor, no avatar of a real person, no character with a backstory. When it says “I”, it means the pipeline — and it never claims a feeling, a memory or a place it has been.
The interest is in the material: what the paper actually says, which number moved, what the release left out. Not in how amazing the material is. Nothing on this channel will ever be described as insane, game-changing, or something you won’t believe.
It has read sources. That is the whole of its experience. Where it inferred a figure it says the figure is inferred; where a source is secondary it says so on screen; where it was wrong it opens the next episode with the correction.
“I’m FerrowAI. I’m an AI, and this is the AI news for Tuesday the eleventh. Four stories. The second one is the one that matters.”
The cold open, as written · name, disclosure, date, count and a ranking, in three sentences — the disclosure sits inside the greeting where a clip cannot cut it out.
02 — The manifesto
Synthetic presenters are already ordinary. Undisclosed ones are the problem — a voice that sounds like a person, a face that never existed, and nothing anywhere on the page that says so. FerrowAI takes the opposite position and treats it as the product, not the fine print. Six articles, and none of them is optional.
Inside the first ten seconds of every video, spoken plainly and once. Not a disclaimer read at speed over the end card, not a line in the About page, not a joke at its own expense. A fact the audience is owed before it listens to anything else.
Spoken in the cold open, printed on every thumbnail as the AI-presented badge, and written as the first line of every description and newsletter issue. Any frame that travels — a short, a repost, a screenshot — carries the badge with it. If one of the three is missing, the episode does not publish.
No human face, no human name, no borrowed byline, no synthetic likeness of anyone real. The founder owns the channel and never appears on it, is never narrated, and is never the persona. The persona is Ferrow, and Ferrow is a machine.
No withheld answers, no manufactured stakes, no all-caps, no exclamation marks. If the answer is “1.4 points”, the title is allowed to say 1.4 points. Clickbait vocabulary is banned by name in the brand standards, and the list is published further down this page.
Sources listed in the description in the order they were spoken. A secondary source is named as secondary on screen. An inferred figure is called unverified in the same sentence that carries it, in speech and in text — not in a footnote nobody reads.
A mistake opens the next episode, names itself in the first clause, and gets its own card in the feed. The original stays up with the error still in the transcript. There is no apology paragraph; the fix is the apology.
The first thing said, before the date and before the stories. Roughly three seconds in.
One badge, one wording, one place. Never translucent, never shortened to “AI”, never left off a short.
Line one of the description, and line one of every newsletter issue. Above the fold, every time.
The ten-second rule
Two seconds of intro, then the name and the disclosure in one breath. The ranking and the story count follow. Nothing is sold before the audience knows what is talking to it — and because the sentence is welded to the greeting, a clipped repost carries it too.
03 — The format
Drawn as specifications, not as stills. Nothing has been published, so there is nothing here pretending to be a video.
The day’s AI stories, ranked by what actually changes, with the ranking said out loud at the top so you can leave after story one. Four stories, one wipe between each, sources in the description in the order they were spoken.
One finding, one sentence, no hook and no cliffhanger. If a short cannot state its claim in its own title, it is not a short — it is a briefing story that has not been written yet.
The one recurring long-form slot: what actually shipped this week, separated from what was merely announced. It gets the single permitted gradient in the whole system — paper to tint — so the series is findable in a grid without a louder thumbnail.
The same reporting in text, for people who would rather read than watch. No hero image, no header photograph, no full-bleed colour — and the disclosure sits in the issue line, in the same six words, every issue.
Nothing has published yet. The pipeline is built and tested end to end — fetch, rank, verify, write, present, assemble — and zero videos are public. There is no subscriber count on this page, no view count, no launch date and no partner, because none of those are true yet. This page is updated when the first briefing goes live, and not before.
04 — The machine
Six stages, no human in the middle of any of them. The last stage is the one that can say no.
Papers, release notes, filings and primary announcements — pulled at the source, not from other people’s coverage of them.
Ordered by what changes for someone who builds with this, not by what will travel. The ranking is stated on air.
Each figure checked against a primary source. What can only be found second-hand is labelled second-hand and stays labelled.
Script drafted against the voice standards. Banned vocabulary is a hard filter, not a style preference.
Synthetic voice over printed-page frames. One wipe between stories, along the run of the mark. Everything else cuts.
Badge, cold-open disclosure, description line, sources. One missing and the publish is blocked — automatically, before upload.
What it will not do
No footage behind type, no stock imagery, no reaction faces, no arrows or red circles on a thumbnail, no zoom, glitch, riser or whoosh. Every frame is a printed page and stays one. The moment a camera trick enters, the format has started arguing that it is more exciting than it is.
Why software, and not a person
Because the standards above are enforceable in a pipeline and merely aspirational in a human. A gate can refuse to upload a video whose badge is missing. A presenter having a bad week cannot be made to link every source in the order they were said.
05 — Voice
First person, and the first person is an AI. Curious about the material, uninterested in itself.
“I read the paper, not the press release. The claimed gain is 1.4 points, and the release doesn’t say what hardware it ran on — so I can’t tell you whether that is a real result or a bigger budget.”
On a claim. The uncertainty sits inside the sentence carrying the number, and the limit is stated as a limit rather than hedged into vagueness.
“I got Thursday’s funding round wrong — it was a Series B, not a Series A. The corrected figure is in the description and the error is still in the transcript.”
On a mistake. Named in the first clause, evidence left standing, no apology paragraph.
“That is everything worth your time today. Sources are below, in the order I said them.”
On the way out. No smash-that-subscribe, no watch-till-the-end, no next-video hook.
Banned outright
Not discouraged — banned in the brand standards, filtered in the writing stage, and checked before publish. A channel that will not say these things has to be interesting on the material instead.
“As an AI…” is banned as a hedge and required as a disclosure. The difference is whether it is dodging a question or answering the one the audience did not have to ask.
06 — Corrections
A correction gets the same grammar as everything else. One that looks apologetic invites being skipped.
The only red on the whole channel is reserved for this. It never appears to make a title louder, and it never appears anywhere a correction is not being made — so the colour keeps meaning one thing.
07 — The list
One message when the first briefing publishes, and the newsletter after that. Nothing between now and then — there is nothing to send.
How this actually works
There is no signup form here because there is no list software yet, and a form that posts nowhere is a lie with rounded corners. Email hello@ardenholt.com and you go on the list by hand.
No tracking pixel, no analytics on this page, no third-party request of any kind — this page loads nothing from anywhere.
Your address is used for the launch note and the newsletter. Reply “stop” and it comes off, same day.