Good morning. Robotaxis are great for cheap rides with zero awkward small talk, but lately they’re causing new problems in San Francisco. Emergency responders have become unpaid roadside assistance when confused Waymos stall in busy intersections. During a recent power outage, multiple robotaxis froze, leaving firefighters and police scrambling to push them out of traffic. Responders even got stuck on hold with Waymo’s customer service, forced to listen to elevator music mid-emergency. No one’s paid enough to deal with that.
Proof we’re (still) not AI Agents: We’re taking Friday off to have some fun with other real humans. See you back here on Monday.
Robots in China now run on OpenClaw. Chinese robotics giant Ecovacs just unveiled Bajie, the first intelligent home butler robot powered by OpenClaw. Instead of rigid programming, Bajie learns your family’s habits over time so it can do the chores you tell it to do. OpenClaw has also been hooked up to Unitree’s G1 humanoid and a robotic arm, allowing people to control them with plain language instead of complex code. It also helps robots understand and remember their surroundings so they can respond naturally to real world context. Meanwhile… anyone seen Optimus lately? [SCMP]
A $50 billion lawsuit could decide who controls AI’s future. Microsoft is threatening to sue OpenAI and Amazon over their new cloud partnership. The fight centers on OpenAI’s latest product, Frontier, and whether hosting it on Amazon’s cloud breaks Microsoft’s exclusive deal. Amazon and OpenAI claim they found a loophole. Microsoft isn’t buying it. Lawyers are now battling over technical fine print, and Microsoft says it’s ready to go public with the fight. If it goes to court, the lawsuit could derail OpenAI’s IPO and trigger unwanted scrutiny. The winner may end up controlling enterprise AI. Probably Claude or China. [Financial Times]
Leads have questions before they buy. By the time you follow up, half are gone.
Chatbase puts an AI sales agent on instantly to qualify, answer questions, and books meetings – without adding headcount.
Trained on your data and live in hours. 10,000+ companies already made the switch.
The Pentagon wants to train AI on classified military intel. The DoD is pushing OpenAI and xAI to feed sensitive battlefield reports and surveillance data straight into their models. Until now, AI answered questions about classified info without actually learning from it. Direct training on top-secret data could speed up threat analysis and target selection. Yet it also risks exposing sensitive details, like operative identities, to unauthorized departments. Hopefully the Pentagon’s AI is better at keeping secrets than the Pentagon itself. [MIT]
China just dropped a self-evolving AI. Chinese startup MiniMax released its new model, M2.7. It now handles up to 50% of tasks researchers typically do, like finding bugs and rewriting code. Instead of waiting for updates, M2.7 learns continuously from its own mistakes. It’s already matching Google’s Gemini 3.1 and approaching Anthropic’s Claude Opus 4.6. MiniMax says their ultimate goal is an AI that fully builds and upgrades itself without human help. Impressive, but it’ll probably take more than self-improvement to lure users away from Claude. Maybe if they open sourced it. [VB]
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 high growth companies.
RunSybil just raised $40M to automate penetration testing. Their AI agent, Sybil, hacks your software exactly like a real attacker, continuously probing apps for vulnerabilities in real time. Founders Ari Herbert-Voss (OpenAI’s first security hire) and Vlad Ionescu (Meta’s former top red-team engineer) argue traditional scheduled pentests and occasional bug bounties are useless against AI-driven hackers. As vibe coding startups ship first and patch security later, AI attackers can exploit holes faster than people can fix them. Your best defense against AI might be attacking yourself first. [Fortune]
Is the era of the solo AI entrepreneur here? Chinese cities are racing to launch 1000s of “one person companies,” turning empty offices and data centers into incubators for AI-powered startups. Local governments offer everything from free workspace and apartments to discounted computing and specialized loans. Solo founders are building entire businesses without employees or investors, relying completely on AI tools like vibe coding agents and automated video generators. Suzhou alone plans 1000 solo AI startups across 30 dedicated communities. By contrast, the US keeps backing big corporations. Maybe startups don’t need teams anymore, just better AI. [Rest of World]
fun stats
👑 73%. Anthropic’s share of spending by first-time corporate AI buyers, up from 50% in January when it was tied with OpenAI. Enterprise AI just got a new king.
🧰 60%. Supply chain disruptions that will be fixed by AI alone by 2031, according to Gartner. Next time gas prices spike overnight, AI will already be on it.
💻 1,800. Lines of code Uber’s AI agent rewrites each week. 84% of Uber engineers eagerly let AI handle entire tasks, no mandate needed.