Good morning. Remember when scratch-and-sniff stickers were peak innovation? Well, now scientists have decided AI needs a sense of smell, too. They’re developing an “e-nose,” and it’s exactly what it sounds like. These AI noses detect odors up to 1,000x more precisely than people and never lose sensitivity. But teaching AI to smell is tricky. Scents are subjective, and there’s barely any data to train on. Can’t wait until Alexa+ suggests deodorant.
Self-improving AI agents boosted their own learning speed by 11%. Former OpenAI researcher Andrej Karpathy recently tested whether AI could independently speed up its own learning. He developed “autoresearch,” a method where agents autonomously design experiments and test changes, then adopt what works. Unlike Google’s AutoML which just automates repetitive training tasks, the Karpathy Loop decides what to test and analyzes the results itself. Shopify CEO Tobias Lutke tried it internally and hit a 19% improvement. If agents can reliably accelerate their own progress, researchers might become optional. [GitHub, Fortune]
OpenClaw is now safe enough for enterprise. Companies love the idea of autonomous AI agents but avoided OpenClaw because it was literally the most dangerous software you could install. Without tight security controls, deploying agents in a business would be a security nightmare. Nvidia finally addressed the chaos at GTC this week by launching NemoClaw, a security wrapper that isolates agents inside containers and strictly limits their access to files and services. OpenClaw launched less than 4 months ago and is already the fastest growing open source project ever. Now that Nvidia’s endorsed it and added real guardrails, get ready for an explosion in autonomous agents. [VB]
Metadata gives your AI context, making it more effective, reliable, and scalable.
In this new guide from You.com, you’ll discover how to:
- Streamline AI projects
- Avoid common metadata pitfalls
- Maximize AI ROI
Set your AI up for success. Get the guide
The Pentagon is vibe-coding its own Claude replacement. Just weeks after banning Anthropic, the Pentagon announced it’s spinning up multiple in-house LLMs. Pentagon AI chief Cameron Stanley says these homemade Claude alternatives will be operational “very soon.” Anthropic spent years and hundreds of millions building Claude, but apparently the Pentagon thinks building cutting edge AI is a weekend hackathon project. And no, this isn’t The Onion. [Tech Crunch]
Mastercard made a $1.8 billion bet that AI agents are the future of payments. It acquired BVNK, a UK startup providing infrastructure for stablecoins pegged to the USD. AI agents can’t open bank accounts (yet), but stablecoins let them transact quickly and cheaply, potentially cutting out traditional payment networks. Mastercard’s new tech lets agents plug straight into its global payments system, enabling seamless digital cash transactions. Analysts call it “future proofing” as more money moves onto blockchain. Now AI agents just need something to buy besides NFTs. [CoinDesk]
Reach your audience
Get in front of 70,000+ tech decision makers with budget and technical influence. 95% US based. 66% open rate. Read daily by AI leaders and builders at growing companies.
If Microsoft ever builds superintelligent AI, it’ll still need to restart twice a week. Microsoft just reorganized its AI teams, putting Mustafa Suleyman in charge of its most advanced models. Until now, Microsoft’s AI efforts have lagged behind competitors. They’re hoping Suleyman can finally deliver the coherent strategy they desperately need. But let’s be real: even if Microsoft’s AI achieves consciousness, it’ll still demand regular updates and the occasional existential BSOD. [WSJ]
Nvidia just wiped out Tesla’s biggest selling point. For years, Tesla’s massive valuation relied on its exclusive self-driving robotaxi tech. Analysts valued that promise at around $1.2 trillion. But at Nvidia’s GTC event this week, CEO Jensen Huang projected a $1 trillion autonomous driving ecosystem built on Nvidia’s DRIVE platform. BYD, Hyundai, and Nissan are already onboard. Even Uber plans to roll out Nvidia-powered robotaxis in 28 global cities by 2028. Nvidia turned Tesla’s special sauce into something any automaker can buy off the shelf. Musk might not be losing sleep over it, but his investors probably should be. [Barrons]
fun stats
☁️ $600 billion/year. AWS sales by 2036 if Amazon CEO Andy Jassy’s AI dreams come true – double his last guess, and 4x what they made last year.
🍺 5,000. Times execs said “AI” on Q1 S&P 500 earnings calls, beating “earnings” by 1,200 mentions. Take a drink every time they say agentic AI? Risky game.
⏱️ 0.03 seconds. How fast AI can analyze customer feedback, 923x faster than a human.