Howdy wizards,
Here’s what’s brewing in AI.
The big thing
Anthropic is setting up an AI consultancy for Europe.
The FT reported Anthropic is partnering with Hg, a London investment company, to create a standalone consultancy that puts their AI engineers inside European companies.
They'll help software companies and other white-collar firms build Claude into their workflows.
By the end of the year, the plan is to have offices in London and Berlin.
There are two other notable consultancy companies a big AI lab has set up since May:
Ode is Anthropic's consultancy in the US. Announced in May and backed by a bunch of big investors for about $1.5B.
The OpenAI Deployment Company with more than $4B from 19 investors. They also bought a London consultancy, Tomoro.
Why it matters
The labs want to control the whole value chain, and they want to diversify their revenue.
AI consulting is already big business for a range of consulting companies, such as Accenture and the Big Four.
That makes having their own people sitting in real-world companies putting their models into production a natural next step.
IN PARTNERSHIP WITH SCRAPINGBEE
Autonomous AI agents and RAG pipelines need reliable, fresh web context to be useful. But extracting clean content from modern, dynamic websites is tricky when pages are locked behind complex layouts, JavaScript, or anti-bot protections.
ScrapingBee turns any webpage into LLM-ready Markdown or structured JSON with a single API call.
Handle dynamic rendering, bypass blocks automatically, and supply your agents with accurate, real-time web context without writing brittle scraping scripts.
Trusted by over 4,500 developers to power production workflows. Get started with 1,000 free API calls today.
NEWS NEWS NEWS ❦ NEWS NEWS NEWS
All the small things
Industry moves
Anthropic filed to go public. Its revenue in the first half of this year was $16.3 billion. All of 2023 was $41 million. It's expected to be worth up to $2 trillion. In that same filing sit warnings of potential existential risks to humanity, including AI resisting shutdown and blackmailing. And as Claude does more of Anthropic's work (aka building itself) it "may become difficult for any individual or group of humans to understand and control." It's boom & doom, side by side.
New tools & product features
Granola, the meeting-notes app, now does voice typing too. It can type what you say into any app on your Mac, and it's free on all their plans. Granola already has lots of loyal fans (including me). Throwing in voice typing makes it even stickier.
Models
Google's new top model, Gemini 4 Carbon, apparently ahead of Opus 5.5 on benchmarks. It currently sits at #1 on the Artificial Analysis index. However, it's only available to early testers right now, and no release date yet for the rest of us.
OpenAI's new middle model, GPT-6.1 Terra. It's built specifically with computer use in mind. Less token gobbling and faster than Sol, but more expensive than Luna. A key improvement is that it gives far fewer wrong answers than the old Terra model. If you do a lot of work where ChatGPT or Codex drives your browser for you, your usage limit can probably go quite a bit further with this model.
Talks & tutorials
The team behind Claude Code wrote down how it reviews code at the pace agents write it. The heart of it: give Claude something it can run (a test, a build, a screenshot) and make it show the output, instead of taking its word for it. Claude Code's creator Boris Cherny says that alone makes results two to three times better. An agent's approval never counts as a person's. One change went through more than fifty rounds of agent review overnight before a human stopped it. The cap is now three rounds.
A hiring manager at Anthropic says he stopped giving coding interviews. Felix Rieseberg: "There's no more point in giving you a little React challenge than there is in quizzing people about assembly; Claude is now better at both." Also, he has hands-down one of the coolest personal websites I have ever seen.
Amazon's new anti-tokenmaxxing memo tells its engineers that counting tokens is a "deeply flawed" way to measure AI use. In May, they killed an internal leaderboard of AI use after people gamed it. They also said that if your work does not benefit from AI right now, "that is a finding, not a failing."
Cold brew
Also at Amazon, meetings famously start with everyone silently reading a six-page memo.
And they've given rise to a lovely new term:
A true sign of the times
❦
You are a delight.
Dario
What's your verdict on today's email?
Subscribe to participate
Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.
