In partnership with
Good morning. A startup says it can turn code into working micro-machines, and its first product is a switch for AI data centers. Also today: the AI lab behind the "model that doesn't hallucinate" is reportedly worth $7.5 billion, and Microsoft shipped a cheap model that only makes decisions.
285 Free Prompts to Save Time
Your inbox isn't the problem. Neither are your meetings. It's the hours you spend on work AI could do in minutes.
285 copy-paste prompts: built for email, meetings, admin, and delegation. Each one has a clear role, a specific task, and a defined output.
No setup. No prompt engineering. Just copy, paste, and get your time back. Free when you subscribe to Mindstream.
TODAY IN AI
1. Atomic Machines left stealth on Oct 7 with $250 million and a system it calls the Matter Compiler. The company says code alone decides what device it builds, with no molds or fixed tooling, and with features down to single-digit microns. Founder Jeff Holden previously helped build Amazon Prime and was Uber's first chief product officer. The first product is the PrimeSwitch PS-150, a 150-amp relay for data centers moving to 800-volt DC. Atomic Machines claims it opens in 50 microseconds, about 1,000 times faster than a conventional contactor. Those figures are the company's own and haven't been independently verified. It is shipping to unnamed early-access customers, and one evaluation product doesn't show the system works across many devices or at scale. Investor Luke Nosek called it "the start of vibe manufacturing."
2. TypeSafe AI, which makes the Jev model, reportedly raised about $870 million at a $7.5 billion valuation, led by Andreessen Horowitz. The figures come from a Bloomberg report on Oct 9 and the company hasn't confirmed them. TypeSafe announced a $40 million seed round only on Sept 15. CEO Diogo Almeida is a co-inventor of ChatGPT. The company says Jev is 20 to 200 times faster and 40 to 400 times cheaper than other models and that it doesn't hallucinate. Those are claims, and the open question is whether Jev's confidence scores hold up in real customer workflows.
3. Microsoft released Microsoft-Decision-1, a small model that returns a calibrated probability for each option you give it, for tasks like routing, classification and checking an agent's work. It is a post-trained Qwen3.5-9B, available in Foundry and OpenRouter at $0.042 per million input tokens, with free output. Microsoft says it is about 35 times faster than GPT-6 Sol and the most accurate across a 36-benchmark comparison. Those results are Microsoft's own tests, and the post was updated Oct 10 with comparisons against JevBench.
FROM THE FRONTIER
Biology gets a shared dataset. At an event on Oct 7, the Zuckerberg-Chan nonprofit Biohub announced $1.8 billion in backing for a "virtual cell" effort. The Department of Energy is putting in more than $500 million over five years, NIH is matching that figure in existing datasets and resources, and Google DeepMind, Isomorphic Labs and Meta are jointly investing $300 million. The goal is an open "world model" for protein structure, protein design and biological discovery that researchers anywhere can use.
The number is softer than it looks. Biohub says much of the value is in data, computing and new measurement technology, and the coverage doesn't itemize how $1.8 billion adds up. No results have been shown yet. Meanwhile Bloomberg reports Isomorphic Labs is in early talks to raise at a valuation of at least $40 billion. The round hasn't closed and its size hasn't been reported.
The reframe. Atomic Machines, TypeSafe and Biohub all sell the same idea: a hard, slow craft turned into something you can request in plain language. The money is large and the proof is thin, with one relay in evaluation, vendor-run benchmarks and a plan. The useful habit is to ask what has been shown and what has only been funded.
See what's really happening in every deal
Aligned shows you what buyers are doing between meetings inside your most important deals then surfaces the risks, openings, and next steps. No more guessing where a deal stands or finding out a champion went quiet too late. Just real visibility, so you can act before it's too late.
IN THE KNOW
Where the relay will debut. Atomic Machines plans to show the PrimeSwitch publicly at the Open Compute Project Global Summit in San Jose, Oct 12 to 15, according to The Next Web. Its website lists no pricing or release date.
Biohub's starting point. The platform draws on an atlas of 6.8 billion proteins and 1.1 billion structures, plus the ESMC language model and the ESMFold2 design engine, per pharmaphorum.
A scanner on a timeline. Midjourney says it is building a full-body ultrasound scanner, with a first scanning site planned for San Francisco around the end of 2027. Its own announcement mentions no FDA approval for diagnostic use, calls its health-outcome figures a possibility, and says the site images are concept renders.
Decision-1's stability test. Microsoft reports that its model's decisions flipped 1.3% of the time on average when inputs were perturbed, with no flips for paraphrased descriptions or reordered options, in the same announcement post.
PROMPT STATION
Find the hard part of your idea. Most ideas fail at one step nobody examined. This prompt finds it before you spend money.
COPY AND PASTE THIS PROMPT
I want to build or launch [DESCRIBE YOUR PRODUCT OR IDEA IN TWO SENTENCES]. Break it into 6 to 10 steps from first customer to delivery. For each step, tell me whether it is easy, solved but costly, or unproven, and why. Pick the single step most likely to kill the project and describe the cheapest test I could run in two weeks to see if it works. Then list three things I'm probably assuming without evidence. Finish with a go, fix or stop recommendation, and the one fact that would change it.
Tip: run the cheap test on the "unproven" step before you build anything around it. If you can't think of a test, that step is your answer.