Three stories this morning, and all three are about where AI actually runs and who gets to check its work: on your laptop, in your company's own stack, and in a little box that tells you whether a picture is real. Let's get into it.
01 of 03
Microsoft Puts Nvidia in a Surface and Gives Windows a Box for Agents
Microsoft finally put prices on the Surface Laptop Ultra, the 15-inch machine built around Nvidia's RTX Spark chip that it first showed off back in May. There are two base models: $2,600, and $3,700 for the one with the beefier chip. Max out the memory and storage and you are at $5,900, and Microsoft says that top model is already out of stock. RTX Spark tops out at 20 CPU cores, 6,144 GPU cores, 128GB of unified memory, and Nvidia's claim of 1 petaflop of AI compute.
There is also a Surface RTX Spark Dev Box workstation starting at $6,000 that ships with VS Code, GitHub Copilot CLI, WSL, and PowerShell 7 ready to go. Dell's XPS 16 Creator Edition on the same chip is $3,800 and up for pre-order at Best Buy, delivering later this month. And if you want to switch, Microsoft will knock up to $1,000 off for trading in a MacBook Pro.
The part I care more about is software. The revamped Windows 11 adds something called Execution Containers, which Microsoft says makes it easier to sandbox AI agents, and Satya Nadella says it is coming to every Windows 11 user, not just people who buy the new hardware. His pitch was that Windows should be a platform "not just for our apps but apps for anyone's agent," and that "just having a model doesn't do much for anything."
Here is where I land on it. Nobody needs a $3,700 laptop to ask a chatbot for a recipe, and I do not think these sell outside of developers and creators with real local workloads. But Execution Containers could matter a lot. Right now letting an agent loose on your PC means trusting it with your whole file system. A real sandbox, built into the OS and handed out for free, is the boring plumbing that makes agents safe enough for normal people to use. I want to see how locked down it actually is before I get excited.
Source: TechCrunch, Microsoft's Surface Laptop Ultra page, and chip specs from Thurrott.
02 of 03
Google Opens Its AI Detector to Everybody
Google launched synthid.com this week, a public site where anybody can upload an image, a video, or an audio clip and check whether it was made with AI. Images can be JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, or GIF. Video can be MP4, MOV, or WEBM. Audio can be WAV, MP3, OGG, FLAC, AAC, or M4A. Until now this was a tool Google had only handed to select journalists and researchers after showing it at Google I/O in 2025.
It works by looking for SynthID, the invisible watermark Google introduced in 2023. Google's own Nano Banana, Veo, and Lyria models stamp it on their output, and so do Gemini, Flow, ProducerAI, and Vids. OpenAI, Nvidia, and Kakao also support SynthID, Apple is reportedly adding support, and Google has built the check into the Gemini app and Chrome. Google says people are already running about 1 million verification requests a day.
What I want to know is what the site says when it finds nothing. A watermark detector can only tell you about content that was watermarked to begin with. Anything from a model that does not use SynthID, or an open model somebody ran on their own box, is going to come back clean. So a "no watermark found" result does not mean "real," and I worry a lot of folks will read it that way. Still, a free public checker that works across several of the biggest labs is a real step. I would rather have a smoke detector that only catches some fires than none at all, as long as everybody knows that is what it is.
Source: TechCrunch.
03 of 03
Nous Research Closes $90M and Takes Hermes to the Office
Nous Research confirmed a $90 million Series B at a $1.5 billion valuation, led by Robot Ventures. Between TechCrunch's reporting and Nous's own post, the investor list includes Nvidia, Microsoft's M12, Samsung, Union Square Ventures, Menlo Ventures, Y Combinator, and 1789 Capital, where Donald Trump Jr. is a partner. The three-year-old company has raised $158 million total.
Nous builds Hermes Agent, an open-source agent it put out under the MIT license in February. The company says Hermes has been cloned more than 24 million times and, by its own estimate, drives about 2.5% of global AI token usage. The Wall Street Journal reported Nous hit roughly $36 million in annualized revenue by mid-September and expects to clear $100 million before the year is out.
The new product is Hermes for Businesses. Companies get customized agents that run multi-step workflows while keeping their data private, plus help picking models by price and performance so they are not overpaying. CEO Dillon Rolnick put it this way in the announcement: "We're building for everyone to own their intelligence, and we're doing it in the open." No pricing yet.
I like this one. Most of the agent money is flowing to closed products you rent by the month, and Nous is making the case that a business should own its agent the way it owns its spreadsheets. The 2.5% token figure is the company grading its own homework, so I would take it with some salt. But 24 million clones does not happen by accident, and an open agent with a real enterprise sales pitch is exactly the kind of competition the big labs need.
Source: Nous Research announcement, TechCrunch.