AI2Day Daily Brief — 11 Sept 2026

Good morning, it is Friday the eleventh of September, and here is what matters in A-I-2-Day dot live have five stories, starting with a sharp argument about the limits of AI confidence, and ending with warehouse robots that sleep less than most software engineers.

Nilay Patel, host of The Decoder at The Verge, published a video essay called People Do Not Yearn for Automation, and it became his most discussed episode in six months, drawing hundreds of comments on YouTube and The Verge. The core argument is worth knowing. AI is genuinely good at writing and checking code because code is verifiable — run it through a compiler and you get a clean pass or fail. The problem, Patel says, is that some people treat that as proof AI can master everything. Medicine, drug discovery, and most of the real world do not have a compiler. Patel links the big AGI claims from the labs directly to this coding strength, and argues the confidence breaks down the moment you leave software behind. It is a measured pushback, and it is worth twenty minutes of your time.

Solid framing for everything else we are about to tell you. On to the defence sector. Swarmer Incorporated, an Austin, Texas company whose software coordinates large fleets of drones and robots, has agreed to buy Kyiv-based ground-robot maker Ratel Robotics for up to two hundred and twenty-four million dollars in cash and stock. The final figure depends on Ratel hitting agreed performance targets. Ratel, founded in 2023, builds unmanned ground vehicles — land-based robots used by Ukraine's military to carry supplies, evacuate casualties, and scout positions. Ratel holds contracts totalling eighty-six million dollars and accounts for roughly thirty-seven percent of Ukraine's Ministry of Defense spending on unmanned ground vehicles so far this year. Swarmer says its autonomy software has logged more than a hundred thousand combat missions in Ukraine since April 2024. The combined company will employ nearly five hundred people across the United States, Ukraine, Poland, and Estonia.

Worth noting the broader money context there: according to Crunchbase, global defence-sector investment hit fourteen-point-six billion dollars as of early June this year, compared with nine-point-six billion for all of 2025. Next, Nvidia. CEO Jensen Huang repeated at the Goldman Sachs Communicopia and Technology conference on Thursday that he expects seventy percent year-on-year revenue growth. Analysts put Nvidia's current fiscal year close to four hundred billion dollars in revenue, so seventy percent on top lands around six hundred and eighty billion. Huang's stated reason for that confidence is structural: almost every organisation building AI right now buys Nvidia hardware and reports back what they are ordering next. He said Nvidia tracks every data centre under construction on the planet. One Nvidia system pairing thirty-six Grace CPUs with seventy-two Blackwell GPUs is growing sales at twenty-seven percent month over month, by the company's own account. Competing chips from Amazon, Microsoft, Google, Anthropic, OpenAI, and startups like Cerebras are all trying to take share. That competition is real, even if Huang is not losing sleep over it yet.

From chips to pure mathematics. Starting in mid-2026, AI models began cracking open problems that had resisted human mathematicians for decades. The headline result: in September 2026, Anthropic announced that Claude helped produce a complete, computer-checked proof of Fermat's Last Theorem — the claim, first stated in 1637, that no three whole numbers satisfy a specific equation involving cubes or higher powers. The proof runs to thirteen million lines in Lean, the programming language used to verify mathematical arguments step by step. Between May and September this year, AI models from OpenAI and Anthropic contributed to at least ten significant mathematical results across number theory, geometry, and probability. Here is the caution, and it matters: several results, including claimed progress on the twin prime conjecture, have not yet been independently verified by outside mathematicians, and some have appeared on corporate websites or social media rather than through peer review. The results are striking. The verification process still has to catch up.

Today's sponsor is Zero2Hero: an AI poker coach sits at your table, and you can ask it mid hand whether the player across from you is bluffing. It runs in your browser on any computer. Windows, Mac, Linux, even a Chromebook. Nothing to install. Lifetime access is $99, in US dollars, until 15 September, at zero, the number two, hero, dot A I.

Finally, Maven Robotics. The Santa Clara startup is coming out of stealth today with a hundred-million-dollar funding round from investors including RoboStrategy, LocalGlobe, Vine Ventures, and XTX Markets Ventures. What makes the backstory worth telling: Maven walked into its first sales meeting in 2024 with no working product — CEO Hamza Derbas described it to TechCrunch as a cartoon of a robot and a team of people — and still beat four rival companies that already had machines on warehouse floors. Today those machines are real. Maven's wheeled robots run up to sixteen hours a day, move at ten miles an hour, lift up to thirty kilograms, and are hitting ninety-nine percent or higher uptime at current customer sites, the company says. They build mixed pallets — the product assortments retailers need but that arrive from factories packed as single-item loads. Maven plans to build two hundred and fifty third-generation robots and begin design on a fourth-generation platform.

If any of those stories deserve a longer look, the Monday edition of AI Today Weekly goes deeper on the week's biggest themes. That is your AI briefing for today. Every story is at A-I-2-Day dot live. That is A, I, the number two, D-A-Y, dot live. We are back tomorrow morning. If you got something out of this, a thumbs up and a subscribe genuinely helps.