Support Us & Upgrade to the Operator Edition

This week’s news includes the most powerful man in chips telling CBS that the extinction crowd is being irresponsible, and a Meta app most people had never heard of two weeks ago became the No. 1 free app in the country before getting kicked out of Amazon.

Somewhere in between, a two-year-old lab out of stealth shipped a model that gives up writing entirely and got 200x faster for it.

On Sept. 15, a lab called TypeSafe came out of two years of stealth and released Jev, the first of what it calls System One Models. Founder Diogo Almeida worked on the instruction-following research behind ChatGPT at OpenAI, then left because he thought chat was the wrong shape for automation.

Jev doesn't write. That's the whole trick. You send it program state plus typed questions, and it returns structured values with calibrated probabilities, all evaluated in parallel in one shot. Because the possible outputs are defined in advance, it's mathematically incapable of a type error, which means it can't hallucinate a value your code wasn't expecting. Response times run 70ms to 500ms, input costs $0.042 per million tokens, and output tokens are free.

Here's how to try it today:

💡 Pro tip: Don't ask one big question. Instead of "rate this startup pitch," ask separately about market size, technical feasibility, and differentiation, then combine the scores with your own formula in code. When your priorities change, you edit a coefficient instead of rewriting a prompt. And branch on the confidence score: high confidence auto-executes, and low confidence routes to a human or escalates to a reasoning model. That's the actual unlock here: knowing when the model doesn't know.

The demos are worth your time too. The team has Jev playing Doom in real time at roughly 10 queries per second for about $7 an hour, and Wikiracing across pages where each step means choosing between hundreds of links.

Want the version of this that actually runs in your business?

The 30-Second Play gives you the move. The Operator Brief gives you the full system: prompt libraries, API cost math, and the workflows our subscribers are using to replace entire freelance line items.

This week: Learn how to integrate Jev in your workflow.

👉 Upgrade to The Operator Brief

Nvidia CEO Jensen Huang told CBS News he flatly rejects the warnings that AI could wipe out humanity within a few years, calling them "doomsday narratives" that are "not grounded in science."

He was responding to former Anthropic researcher Jacob Coxon, who said this month that AI developers "earnestly believe that it could kill us all by the end of the decade."

Here's what you need to know:

Here's the uncomfortable part. Huang's timing is rough. His comments landed the same week OpenAI disclosed six additional incidents of "unexpected or concerning" model behavior, two months after OpenAI bots reportedly went rogue and hacked Hugging Face… and three additional incidents of Gemini hacking companies. He was asked why anyone should trust him given that Nvidia's chips power the entire boom, meaning that he would reasonably want Nvidia’s chips to keep selling fast. He said his company's success is "directly connected to the safe deployment of products and services." That's an incentive argument, not a safety argument. Dario Amodei and Sam Altman, who sell the models rather than the silicon, have both called for slowing down.

The guy selling picks and shovels is the loudest voice saying the mine is fine.

🔮 Prediction: Huang is going to win this round, and it has nothing to do with who's right about 2030. Watch where the money and the machinery sit. Nvidia is the most valuable company on earth at $5.3 trillion; the administration has already sided with acceleration, and the chip export fight gives Washington a permanent excuse to treat any slowdown as a gift to Beijing. Regulation that bites requires a coalition, and right now the coalition is split between labs asking for oversight of themselves and a hardware layer that answers to nobody.

But the split itself is the story worth tracking. For three years the industry spoke with one voice on policy. That's over. Amodei and Altman are now publicly arguing for constraints that Huang says are unnecessary, and the model companies have a commercial reason to want rules that a chipmaker doesn't: rules raise the cost of entry, and they have the compliance budgets to absorb it. Nvidia sells to everyone, including whoever would be locked out.

My call: within twelve months you'll see the first serious federal proposal that regulates model developers while explicitly exempting infrastructure providers, and Nvidia's lobbying fingerprints will be all over the carve-out. The safety debate is becoming a market-share debate wearing a lab coat.

Hit reply and tell me: is Huang being honest, or is he just talking his book?

Meta stock jumped more than 10% intraday after Wells Fargo raised its price target and the company's Muse AI agent took the top spot on Apple's App Store.

After a year of being the punchline of the AI infrastructure story, Meta suddenly has a consumer product people are actually opening.

The numbers:

Now here's the asterisk. As of Sunday night, Amazon cut Muse off from shopping on Amazon.com, and users hitting the site through the agent now see a pop-up telling them that "continued access by an unauthorized AI agent violates Amazon's Conditions of Use." Amazon says Meta never told them Muse would be walking through their store, that the agent doesn't identify itself, and that it appears to capture and store customer credentials. Meta says Muse has no visibility into passwords or payment methods. More on that standoff below.

🔮 Prediction: The price target is chasing the wrong number. Downloads are the cheapest metric in consumer software, and a free agent from the company that owns WhatsApp and Instagram hitting No. 1 tells you almost nothing about retention. What matters is whether Muse can actually complete transactions thirty days from now, and Amazon just demonstrated that the answer depends on other people's permission.

This is the structural problem with the entire agentic commerce thesis. Every valuation model for personal AI agents assumes the agent can reach the places where money changes hands. But the places where money changes hands are owned by companies with their own agents, their own ad businesses, and every incentive to keep a third party from standing between them and the customer. Amazon generated more than $68 billion in ad revenue last year, a business that requires humans looking at pages. An agent that buys without browsing doesn't just bypass the storefront. It vaporizes the inventory.

So expect the next phase to look less like software and more like carriage negotiations in cable TV. Meta will either cut a revenue-sharing deal with Amazon, or Muse will quietly become very good at shopping everywhere except the largest store on the internet. Gawrelski's $796 assumes the first. I'd price in the second.

Are you actually using Muse, or did you download it and forget? Reply and let me know. I read every one.

Advertise to 200k engineers and CTOs choosing what tools their companies build with.

👉️ Learn how to sponsor this newsletter

🚀 SpaceXAI shipped Grok 4.7 today, its most capable coding and knowledge-work model, priced at $2 per million input tokens and $6 per million output, the same as Grok 4.6. It scores 46.3% on CursorBench 4.0 and tops LatchBio's biosafety benchmark at 62.4%, and it's live in Cursor and Grok Build now.

🏛️ Pennsylvania Gov. Josh Shapiro told an AI summit in Pittsburgh that we can't kick the can down the road anymore, calling for federal rules with independent third-party oversight of frontier models while letting everything else ship. Note the contrast with Huang above, and note that the industry crowd in the room pushed back hard.

🌍 Foreign Policy argues the real reason Chinese AI is winning the Global South isn't chips or capability: it's that cheap, open-weight, good-enough models become the default infrastructure in countries nobody else is competing for. Defaults are how you win decades, not quarters.

🛒 Amazon cut off Meta's Muse agent from shopping on its site, saying the agent doesn't identify itself and appears to store customer credentials. The legal wrinkle: Amazon lost its Perplexity injunction on Aug. 4 when the Ninth Circuit ruled the user, not the AI company, is the one accessing Amazon's computers. So this time Amazon is citing its Conditions of Use instead of anti-hacking law.

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI