Seven stories this week and a few small finds at the end. Each one happened out there. If you lead a small team or a small newsroom, you will see yourself in some of it. I try to leave the connecting to you.

Who gets to say it is settled? OpenAI says its agents solved a million-dollar maths problem. The prize committee says "apparently". A software company's board voted on a Wednesday, and by Saturday the company said the founder was back. One wiki host says nine in ten new wikis are shut out of Google, and calls its own explanation a guess.

Who is visiting? A media chief executive says agents are coming to her sites and the young readers are not. A software founder says his docs traffic fell by about 40 percent while the software was used more than ever. A search engine seems to decide that a new site is not worth showing.

How small can the first step be? A handful of people in a walled-off corner of the office. Three thousand desks out of fifty-four thousand. Each of them is a way to try something large without betting the organisation on it.

A note on how I write this. I mark what is confirmed, what is reported, and what is my own reading. Where a source is a company talking about itself, I say so.

In plain words: these equations describe how water and air move. The question is whether a smooth flow can ever develop a point of infinite speed. The Clay Mathematics Institute put a one-million-dollar prize on it in 2000. Its problem statement has four versions, A to D. A and B have no outside force. C and D allow a smooth force.

What is reported, not confirmed: the agents reached the result on Saturday 5 September, about 88 hours after the run began, and Lean verification took 17 more hours. I could not open OpenAI's blog, so those figures come from secondary coverage.

The same day, a declaration called "A Severe Misalignment of AI in Mathematics" appeared. It says: "The goals of the AI companies and the goals of the mathematical community are severely misaligned." Terence Tao, Peter Scholze and Pierre Deligne are among the signatories. The number of Fields medalists differs between the Zenodo record and the declaration site, so I say more than twenty.

By the end of Saturday 12 September, the prize committee's own word was still "apparently".

On Wednesday 9 September, Automattic's board put its CEO on paid leave. The company's statement: "Matt Mullenweg is currently on leave from Automattic. Mark Davies, Automattic's CFO, will lead the company as interim CEO."

By his account, four voted for and he voted against. The four-to-one count is my arithmetic from his message, not a published tally. He also says he asked repeatedly for a few hours with independent legal counsel and was refused. No reason for the vote was given publicly.

So a vote on Wednesday was gone by the weekend. What the record cannot tell me is why. I found nothing on who owns what, who appoints the directors, or what the votes weigh. Neither side has explained.

My reading: a vote starts a leadership change. How it is run decides whether it holds.

The post is dated 17 August. It took off this week, after it hit Hacker News on 8 September and reached 575 points. The author, cookmeplox, says that for two years wikis on brand-new domains "will not show up on Google Search for anything other than the main page."

Read the 90 percent for what it is. It is a tally of wikis he knows about. The post gives no sample size and no list. He gives his explanation as "a guess": that Google gave up on spotting SEO slop and decided to nerf brand-new domains.

I found no Google comment as of 12 September. Google's March 2024 post introduced three spam policies and says nothing about a new-domain penalty.

The 85 percent of game-wiki traffic the author says comes from Google is his own estimate, for game wikis only. A new local news site is a different case, and I gathered no evidence from news sites. Treat that link as my hypothesis.

The author ends by asking anyone with data to contact him. The data is not in yet.

Siv Juvik Tveitnes said that on the INMA podcast "No One's Reading This", published 10 September. It is not her own line. She introduces it as "a very good saying". Tveitnes is CEO of Schibsted and became chair of INMA, the media industry body, on 1 June (a date from a secondary source).

She gives no number and no share of traffic. It is an observation without data, and she says so by not offering any.

The wider data points the same way, but it does not measure her claim. The Reuters Institute Digital News Report 2026, published 16 June, says 52 percent of 18 to 24 year olds name social media or video networks as their main news source.

None of that counts agents visiting websites. Only she says that part. One caveat on my side. The transcript is automatic, with names misspelled. I have not checked the words "agents" and "never arrive" against the audio.

Ugarte leads product for Grok Bot at SpaceXAI. The goal, in his words, was to build a knowledge work product that brings agents to the rest of the company.

The cave was real. The team sat in a separate part of the office, with private Slack channels. His reason: they had to make many small decisions every day.

Now the discount. Nearly everything about success comes from Ugarte, SpaceXAI or the host. Lenny calls Grok Bot "the hottest AI product in the world", says he is obsessed, and the episode has ad breaks.

Nobody has published how many people were in the cave.

In December 2025 the federal administration said it had finished rolling out Microsoft 365 on about 54,000 workplaces. On 2 September the Federal Chancellery launched a programme for a "digitally sovereign workplace". From the end of 2027, around 3,000 employees are to use it. The estimated cost is 9 million Swiss francs, for the first phase.

The platform is openDesk, run by Germany's ZenDiS. It covers mail, calendar, documents, presentations, telephony and video conferencing, and it runs in parallel with Microsoft 365. "Pilot" is the press's word. The chancellery says programme.

Some coverage wrote "replacing Microsoft". The feasibility study says otherwise: full substitution "is not a realistic option and currently comes with too high risks." The use it names is an emergency fallback if today's office software fails, and a place to handle sensitive information.

The study, which I read directly in German (quotes below are my translation), ran a user test from December 2025 to March 2026. Capacity was 300 users. 172 testers from several departments signed up. The browser solution averaged 5.4 out of 6.

It says the results are a basis for deeper work "but not yet ready for production". It also says: "The PoC delivers no reliable economic calculation for going into production." The 172 were registered volunteers testing a browser version, which may flatter the grade. That is my caution, not the study's.

The study's own last pages say the next step has not been decided.

On the business: "we'll no longer be trying to grow the business around Tailwind." Existing customers keep their access to Tailwind Plus and ui.sh. Sign-ups for new customers are closing. No financial terms are public, and "joining" may or may not mean an acquisition, so headlines saying "acquires" are an interpretation.

The post mentions AI only in passing, as Shopify's work on agentic commerce. It says nothing about docs traffic or layoffs. For that you need January.

He explained that the docs were the only way people learned about the paid products. I have those quotes through secondary summaries, and the layoffs date to early January. DevClass reports three engineers. The 40 percent is his number, with no independent traffic data behind it.

Secondary outlets tie January to September as cause and effect. Wathan does not say that in the September post. The counterpoints are fair. The analyst paddo.dev argues the problem is a business built on traffic, not a dying framework.

My own inference, labelled as such: if people ask an assistant instead of reading your pages, the pages that sell are the ones you lose first.

That is the issue. Mattias