Howdy wizards,
Here’s what’s brewing in AI.
The big thing
SpaceX is buying Cursor for $60 billion, exercising the acquisition option it took out back in April. This comes right after SpaceX debuted on Nasdaq with the biggest IPO in history.
Cursor isn't just a massively popular AI code editor (≈$1B ARR), it's also building its own frontier model. They recently announced Composer 3, a model built from scratch, in the same class as Claude's Opus. It was trained on 100,000+ GPUs from Musk's own xAI, and Cursor says it will release it in the coming weeks.
They’ve also just launched Origin, basically a version of GitHub for the agentic era. It does things like automatic merge-conflict resolution.
Why it matters SpaceX wants to own everything: the model, the editor, and the place the code lives.
Cursor went from a very convenient wrapper to now building its own frontier model on a gargantuan cluster of compute, and going head-to-head with GitHub to own the full lifecycle of code.
Musk’s AI company already had the compute. Now they’re getting Cursor's coding data, customer list and talent — all things that make them a credible frontier contender alongside Anthropic, OpenAI and Google.
IN PARTNERSHIP WITH GRANOLA
You know when you’re in a meeting and someone’s random bot joins the call?
With Granola, you get exactly zero meeting bots joining your call.
It uses your device’s own audio to transcribe and works with any meeting tool (Zoom, Google Meet, Microsoft Teams, etc). It even works for in-person conversations.
You stay focused and jot down notes like you normally would. Granola quietly transcribes and enhances the important bits in the background.
If you want to be extra thoughtful (or your company requires it), Granola can auto-send a quick consent email beforehand.
Try Granola on your next meeting and see how much easier it is to stay present.
NEWS NEWS NEWS ❦ NEWS NEWS NEWS
All the small things
Industry moves
Anthropic's two best models are still banned, and security researchers are pushing back. Last week, the US government forced Anthropic to switch off Fable 5 and Mythos, its most capable models, on national-security grounds. More than 100 security researchers have since signed a "Free Fable" open letter arguing the ban does the opposite of what it intends: it stops defenders from using these models to find and fix security holes, while attackers (who don’t care about US export rules) could just switch to the best model available, even if they’re e.g. open Chinese models that can’t be banned.
DeepSeek is now China's most valuable AI startup. In its first-ever round, it raised $7.4B and reached a $50B valuation. Read it together with the open-weights releases below: the cheap, open end of the market is increasingly Chinese.
42 states subpoenaed OpenAI over how ChatGPT behaves. Authorities are now demanding records on ChatGPT's ads, engagement tactics, data handling, and treatment of minors, with sycophancy named as a concern.
New tools & product features
Midjourney did something surprising: they built a body scanner. The company known for being a text-to-image generator unveiled Midjourney Medical, a 500,000-sensor ultrasound rig that maps your whole body in about 60 seconds, no radiation, a few dollars a scan. The first "Midjourney Spa" opens in San Francisco in 2027, with FDA approval pending.
Claude Design now stays on-brand more reliably and hands off to code cleanly. Anthropic's prompt-to-UI tool could already reference your brand's design system, but the update makes it more precise: it now checks its own output against it and fixes off-brand mistakes before you see them. A new command moves a design into Claude Code and back, and it uses fewer tokens.
Models
China just shipped the most capable open model yet. Z.ai's GLM-5.2 is free for anyone to download and run on their own servers. On coding it reportedly matches Anthropic's and OpenAI's best models, at a fraction of the cost. A second Chinese lab, MiniMax, released a comparable model the same week. Top-tier coding AI is no longer something you can only rent from a US lab. Soon, it’ll start squeezing what OpenAI and Anthropic can charge.
Research
New Anthropic research shows understanding the problem beats having a coding background. Anthropic went through 400,000 Claude Code sessions and found most non-technical people who were knowledgeable in their domain managed to ship working code almost as often as professional engineers. The key finding is: domain expertise is the main thing that amplifies the effectiveness of a tool like Claude Code.
❦
You are a delight.
Dario
What's your verdict on today's email?
See which AI use cases are paying off with Context Windows Pro
Most companies pick AI use cases by brainstorming internally. 90% of those initiatives fail.
I’ve created Context Windows just so you can pick the winners.
🟦 Find high-performing use cases from 2,000+ companies at contextwindows.ai, or book a demo with me
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.