What Gets Into This Brief, and What Gets Cut

Call this issue zero. No news in it. Just the rules, so that when the first real edition lands you already know what the three slots mean, what we refuse to print, and what happens on the morning we get something wrong. Five minutes now, saves you guessing later. ☕

Why this brief exists

AI news is produced faster than any working person can read it, and most of it is not news.

It is announcements. Waitlists. Demo videos edited to the second. A model scoring well on a test the vendor chose. Somebody's thread about somebody's blog post about somebody's announcement. By the time it reaches you it has been through four hands and nobody is quite sure what the original document said.

Meanwhile the thing you actually needed to know, that a tool your team already pays for quietly changed what it does with your files, goes out as a help-centre note and nobody picks it up.

This brief is built for one reader: somebody who runs or works in a small team, has a real job that is not following AI, and needs to know which of yesterday's noise changes their week. Not the industry's week. Theirs.

Which means most mornings the honest answer is that very little changed, and we would rather say that in one line than pad the page. A short brief is not a lazy brief. It is a brief that did the cutting before it reached you.

Three slots, every morning

Same three slots, same order, every weekday. The shape never changes, so you learn where to look and stop reading the parts you do not need.

• What actually shipped. Something that exists and that a reader could use today. Named, with a link to the vendor's own page. A thing in general release and a thing behind a waitlist are different animals, and the line says which one it is.
• What it costs. Free tier, paid tier, usage-metered, or not published. If the vendor has not published a price we write that the price is not published. We do not estimate, and we do not repeat a figure we saw in another newsletter.
• Who it changes the week for. The honest answer, including when the honest answer is nobody yet. If a release only matters to teams with an engineer to wire it up, the line says so, and you get your morning back.

The one-primary-source rule

Nothing enters this brief without a primary source we can link.

A primary source is the document from the people who did the thing. The vendor's own announcement. The changelog entry. The pricing page. The terms or privacy page. A regulator's filing. A court docket. A research paper on the archive where it was posted. A transcript of someone saying it themselves.

If the only thing we can find is an article describing a document, we do not have the document. We have somebody's reading of it, and readings drift. A nuance drops. A conditional becomes a certainty. "Rolling out to some customers" becomes "launched".

So the rule is blunt: find the paper, link the paper, or cut the item. There is always another item.

One consequence worth stating. This brief will sometimes be slower than the fast accounts. Something will circulate for a day before it appears here, because the underlying document took a day to surface. We think a reader who gets it a morning later and gets it right is better served. If you need to be first, this is not your newsletter, and we mean that kindly.

Four things that get cut

Most of the editing happens before anything is written. Four categories go straight in the bin, no debate. ✂️

• Stories with no traceable origin. A screenshot, an anonymous claim, a post that says a source confirmed it. Interesting, unprintable. If we cannot get back to the thing itself, it is not an item.
• Coverage of coverage. An article about an article about an announcement. Three layers of paraphrase, each one a little more confident than the last. We go to the announcement or we skip it.
• Press releases in a trench coat. Vendor language lifted whole, with its verbs intact. Revolutionary, seamless, industry-leading. If an item can only be described in the words the vendor chose, it has nothing in it yet, and we say what shipped in plain words or we leave it out.
• Numbers with no paper trail. Market sizes, adoption rates, benchmark results, user counts quoted with no document behind them. A number that cannot be traced is worse than no number, because a number gets repeated, forwarded and put in somebody's board deck.

Announced is not shipped

Half the confusion in AI coverage is one missing word. A thing was announced, and the sentence came out as though you could go and use it.

So every item in this brief says which stage the thing is at, using the same five words every time. The vendor's own page is where we get it from, and when the page is vague we say the page is vague.

• Announced. A blog post exists. The product does not, or not for you. Useful to know, not useful to plan around.
• Waitlist. You can register interest. There is no date, and a waitlist with no date is an announcement with a form attached.
• Limited. Live, but only for selected accounts, one region, or one plan tier. The interesting question is which tier, and we ask it.
• Generally available. You can sign up this morning and it works. This is the only stage where the phrase "it shipped" belongs.
• Going away. A deprecation, a sunset date, a plan being retired. Least covered category in the whole field and usually the one that costs a small team actual money.

Who writes this

A desk, not a persona. There is no invented correspondent with a friendly first name and a made-up biography, because an audience that trusts a character is trusting something that cannot be held responsible.

So items appear under the desk's name. When a human judgement is being made rather than a fact reported, the wording makes that visible: a call, a guess, a thing we are unsure about. You should be able to tell the difference between the brief telling you what a document says and the brief telling you what it thinks.

Software helps assemble the brief. It does not get the last word, it does not get to pick the three slots, and it never supplies a fact that a person has not traced back to a primary source. Anything that cannot be traced does not run, no matter how plausible it sounds, and plausibility is exactly the failure mode worth guarding against.

The brief has no newsroom, no insiders and no scoops. What it has is a reading habit and a short list of rules it is willing to be judged on. That is the trade: less access, fewer claims, and every claim checkable by you in one click.

Why a roundup blog is not a source

This is the rule people push back on, so here is the reasoning in full.

Aggregator posts are useful. We read a lot of them, and they often point us at things we would have missed. But a roundup is a reading list, not evidence, and treating it as evidence introduces an error we cannot see.

The failure is cumulative. Somebody summarises a changelog and trims a qualifier for space. The next writer picks up the summary and adds a confident headline. A third writer cites the headline. By the fourth hop, a limited release to selected enterprise accounts has become a general launch, and nobody lied at any step. Everyone just compressed, in the same direction, because compression always drops the caveats first. Caveats are the boring part.

There is also a volume problem. Roundups reward completeness, so they list everything. This brief rewards omission. If both of us put the same twenty items in front of you, one of us has not done our job, and it is not supposed to be us.

So aggregators get used the way a tip line gets used. They tell us where to look. Then we go and read the document, and the document is what we link.

What happens when we get it wrong

We will get things wrong. The useful question is what lands in your inbox the next morning.

Corrections go at the top of the next edition, not at the bottom, and not tucked into a web page you would have to go looking for. The note says what we printed, what was actually true, and where the mistake came from. If an item was wrong enough that acting on it would have cost you something, the correction leads the edition.

The archived edition gets a dated note too. We do not quietly fix the old page and move on, because a silently repaired mistake looks exactly like a mistake that never happened, which is the whole reason it is tempting.

And we distinguish two kinds of wrong. Getting a fact wrong is an error and gets corrected. Reading a situation wrong is a bad call, and those get revisited out loud rather than buried. If this brief says a release does not matter and three months later it plainly does, you will read that sentence here.

Nobody has to ask nicely to get a factual correction. Corrections are not favours.

Words you will not read here

Hype is a style before it is a claim. Cut the style and a surprising amount of the claim goes with it. The following are banned outright, in headlines and in body text.

• Game-changer, revolutionary, paradigm shift. Verdicts dressed as adjectives, delivered before anyone could possibly know.
• Shocking, insane, wild, breaks the internet. A description of how we want you to feel, standing in for a description of what happened.
• The end of search, the death of coding, the last job. The obituary genre. Expensive to believe and never retracted.
• Quietly, secretly, nobody is talking about. A release published on a company blog was not quiet. It was just not read.
• You need to see this. No you do not. If it matters we will tell you why in one line, and if we cannot, that is our answer.

Reading an edition in ninety seconds

The format is fixed so you can skim it in a predictable order and bail early without missing anything that was meant for you.

Start at the top line. One sentence saying whether this was a quiet morning or a loud one. On a quiet morning you can stop there, and plenty of mornings are quiet.

Then the three slots in order. Each one opens with the thing that happened, in plain words, before any interpretation. If the first clause does not concern you, skip to the next slot. That is the design, not a failure to engage you.

Every named tool links to the vendor's own page, so a link is always a way to check us rather than a detour. Anything we have written about at more length is linked too.

At the bottom, one line on what we are watching and expect to cover when there is a document to point at. Not predictions. Just the things with a date attached. Our full editorial policy lives on its own page, and it is the version that wins if this note ever drifts from it.

Sending us something

Tips are welcome, and the fastest way to get one used is to include the link to the thing itself.

A tip with a primary source can be checked in minutes and may run the next morning. A tip without one goes into a queue where somebody has to go and find the document, and most of those never come back out. Not because we doubt you. Because the day is finite.

Founders and PR teams are welcome to send things, with two conditions. Tell us what is actually available today, as opposed to announced, demoed or planned. And tell us the price, or tell us there is not one yet. Both are printable. A pitch that will not say which stage the product is at answers the question by refusing it.

We take no payment for coverage, ever, in any form. Some links elsewhere on the site earn a commission, which is disclosed on the page where it happens, and no commission has ever made the three slots. The about page spells out the arrangement.

Embargoes get read, not signed. If a condition is that we agree in advance to a framing, we will wait and write about it when everyone else can.

What this brief does not do

Worth being clear about the gaps, so nobody waits for something that is not coming.

It does not predict. No forecasts about which company wins, no guesses at unannounced products, no commentary on share prices. Nobody knows, and the people who sound most certain are selling the certainty.

It does not review. A one-line mention that a thing shipped is not a judgement that it works. Where this site has done the longer reading, the item links to it and labels it.

It does not cover the industry as a sport. Funding rounds, executive moves and conference drama change nothing about your Tuesday, and that is the test every item has to pass.

It does not pretend to first-hand use. If an item is based on a document rather than on somebody opening the product, the wording says so. There is no sentence in this brief implying a trial that did not happen, and there never will be.

Three things, one source each, no adjectives doing work the facts should do. Starting tomorrow morning.