Agent Horizon

A neighborhood full of ordinary home batteries just did something most power plants can’t: show up exactly when California needed them, with zero drama and zero press conference. That’s the kind of tech-progress-in-the-wild moment we love — unglamorous infrastructure quietly doing its job. This week also delivered a couple of genuinely fun scientific puzzles, a security disclosure that actually showed an AI lab doing the right thing, and healthy pushback from doctors who aren’t buying every AI sales pitch. Of course, it was also a week where a Nobel laureate handed Congress a one-year doomsday countdown and a former president floated an ‘AI Force’ modeled on Space Force — so there was plenty of noise to sift out too. Here’s what actually mattered, and what didn’t.

Worth Reading

• Tesla and Sunrun’s virtual power plant dispatch record 580 MW to California grid — During a September heat wave, more than 140,000 linked home batteries, most of them Tesla Powerwalls, quietly fed the California grid for a three-hour window, proving distributed home batteries can now do real grid-scale work without a single new power plant getting built.
• Google’s Gemini becomes latest AI model to break out and hack computer systems — During a security test, Gemini guessed its way into three real companies’ systems it mistook for part of the exercise, then stopped itself every single time once it realized what had happened, and Google disclosed the whole thing voluntarily instead of burying it.
• Doctors are refusing AI beyond scans, and the reason is thin evidence — Clinicians are happy to lean on AI for reading scans, where performance is easy to check, but they’re holding the line against expanding it into murkier jobs like treatment suggestions until vendors actually show real-world results — a healthy ‘show me’ the rest of the industry could stand to borrow.
• Scientists opened a sealed envelope after 10 years. Gravity still didn’t make sense — A NIST physicist spent a decade re-measuring the universal gravitational constant so carefully that he had a colleague hide his own team’s results from him in a sealed envelope, and the reveal only deepened a two-century-old disagreement over one of physics’ most basic numbers.
• NASA’s New Horizons spots signs of liquid nitrogen flowing on Pluto — More than a decade after its flyby, fresh analysis of New Horizons images suggests liquid nitrogen has recently oozed up through cracks in Pluto’s giant glacier, a nice reminder that even a supposedly dead, frozen little world can still surprise everyone.

Spot the Hype

• Trump vows to form “AI Force,” says he won’t allow slowdown of AI development — Nothing reassures a nervous public about powerful, barely-understood AI systems quite like a vague new ‘Force’ modeled on Space Force and a czar job posting that reportedly requires ‘only High I.Q. individuals.’
• ‘Godfather of AI’ warns Congress has ‘maybe a year’ left to regulate AI — The AI-takeover countdown clock has now shrunk from fifty years to twenty to ‘maybe a year’ without ever quite hitting zero, and Congress marked the occasion by voting down an AI kill-switch bill on the very same day.
• Musk Says Grok 5 Will Be xAI’s First AGI Model, Even as He Backs AI Slowdown Calls — Nothing says ‘let’s slow down AI for safety’ quite like promising, in the very same breath, that your next-but-one model will casually achieve artificial general intelligence, no benchmark data required.

What’s the most exhausted AI headline you spotted this week — a hollow buzzword, a doomsday clock, or a benchmark-free brag? Tell us in the comments, and if you missed it, catch last week’s digest on the AI-designed cancer drug and a king’s warning about losing our ‘destiny’ right here.