Agents Are Moving Into the Real World

September 24, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
September 24, 2026

Take our listener survey to help shape the direction of the show.

HEADLINES

Anthropic's Biolab Claims Its First Discovery

Less than a week after confirming it runs an AI-powered wet lab in the Bay Area, Anthropic says Claude discovered a novel enzyme system with CRISPR-like repeats "with only high-level direction from our scientists." Anthropic admits it doesn't yet know what the system does, but the search took roughly 1,000 Claude agents, 21 hours and 210 million tokens, somewhere on the order of $10,000 of compute. Dario Amodei framed it as the start of the same weak-to-superhuman curve AI followed in math, and a first step toward the "cure most diseases in 5-10 years" goal from Machines of Loving Grace. Critics called it overhyped: genome mining experts note that finding a weird cluster of genes and repeats is the easy part, and figuring out what it does is where the real discoveries come from. The fair summary: very cool, and very, very preliminary.

Trump Renames AI "Super Intelligence" at the UN

In a speech rejecting "any attempt to construct a globalist scheme to control Artificial Intelligence," President Trump declared the technology "hereinafter officially called Super Intelligence," because "the use of the word artificial makes it sound fake." The memes were infinite, and AI safety advocates groaned at the namespace clash with actual superintelligence. The substantive objection didn't come out of the blue: a Finland-led group of 20 countries signed a declaration calling for mandatory independent safety testing and an international AI institution, backed by UN Secretary-General António Guterres, with neither the US nor China on board. White House science advisor Michael Kratsios doubled down at the Security Council: "A prosperous future will not be secured by a global regulator."

Axios Trump orders AI rebrand as "super intelligence"
The Verge Trump says the US is officially renaming AI to 'super intelligence'
Bloomberg Trump Rejects Push for International Pact to Manage Risks of AI
The Information AI, the UN, Xi and Trump
WSJ WSJ Opinion: Donald Trump's U.N. Speech on Iran and 'Super Intelligence'
Aaron Rupar (X) Video: "hereinafter officially called Super Intelligence"
Aaron Rupar (X) Video: "We will only encourage super intelligence, not rein it in"
Office of the President of Finland A Call for Control of Frontier AI Models
United Nations Statement by the Secretary-General on Artificial Intelligence
The Verge World leaders are backing frontier AI controls, but the US and China are nowhere to be seen
Bloomberg UN Chief Calls for AI Pact to Prevent Rise of Killer Robots
Michael Kratsios (X) Advancing frontier is no reason to pause or add global governance structures
Amjad Masad (X) "Your intelligence is fake... Mine? It's super."
Zvi Mowshowitz (X) Of all words, "super" is a linguistic nightmare of a namespace clash

Altman and Amodei Brief the UN Security Council

Sam Altman, appearing live in New York, built his remarks around AI's dichotomy and two risks: recursive self-improvement producing models labs can't understand or control ("It doesn't matter whether people put their risk of catastrophe at 10% or 1%, none of these levels are remotely acceptable"), and concentration of power in any company or country that believes only it can be trusted. Dario Amodei, by video link, was far less coy on governance, calling AI "the most important global security issue facing the world today" and laying out a playbook: simple treaties like a ban on AI-developed bioweapons, global evaluation and verification systems, and shared testing standards with a global notification system. The result is a clear map of the discourse: Anthropic on one pole, the White House on the other, and OpenAI trying to chart the middle path.

MAIN STORY

Personal Agents Move Into the Real World

The debate over whether consumers actually want personal agents is starting to get real evidence. Muse sits at number one on the App Store, a NYT reviewer came away blown away, and at Meta Connect the agent moved off the screen and onto glasses and a new pocket device, while GrokBot landed in Teslas. Skepticism hasn't gone away, but the people actually trying these products keep coming away more impressed than they expected.

NOBODY WANTS TO BE PRODUCTIVE

The Case Against Personal Agents
"People love shopping. They don't want to be productive."
Stratechery's Ben Thompson went viral on TBPN arguing that booking flights is a perpetual demo nobody wants, that shopping is entertainment, and that people enjoy complaining about errands more than they want them gone. StaySaaSy argued agents don't abstract decisions away, they multiply them, which is a nightmare for everyone except type-A strivers. Kurt Vonnegut made the romantic version of the point in 2005: walking out to buy a single envelope is the fun part, because "we're dancing animals."

The Counterargument
"That's not shopping, it's buying things."
Painting all consumers with one brush is reductive. Some people hire stylists and travel agents to save time; everyone has schlep work like booking a kid's class or coordinating a dinner that is the opposite of entertainment. Browsing clothes can be fun while hiring a roofer is miserable, and the same person can want both experiences on different days. The latent demand for agents that remove that friction is immense.

The Evidence
"The most useful A.I. app I had ever used."
Muse continues to hold the number one spot on the App Store, and it's getting positive buzz from unexpected places. The NYT's Eli Tan handed his life to Muse and found it filled out forms, paid a parking ticket, called his dental insurer and even bargained with Facebook Marketplace sellers on his behalf. Coming in, his skepticism was about whether agents were useful enough; coming out, it was about whether enough people will be comfortable with the privacy bargain.

Meta's Partnership Blitz
Grabbing the momentum by the absolute throat.
Meta clearly believes it has something, with chief AI officer Alexandr Wang on an X frenzy announcing partner after partner, including Spotify, Box, Sephora, Gap and Walmart. New features landed too: computer use in the Muse for Mac app and a real-time avatar that lets Muse talk back to you in video form.

Meta Connect and Muse Charm
A Steve Jobsian "one more thing" for your agent.
After a few anticlimactic years, Meta Connect was built around bringing Muse into the real world: a Ray-Ban refresh with the first camera-free audio pair, and Muse on glasses for hands-free conversation. The finale was Muse Charm, a palm-sized, always-on device with a small color screen that Zuckerberg called "by far the fastest way to talk to your Muse" for anyone without Ray-Bans; it's still in active development, with Meta hoping to ship by the holidays.

The Reaction
Six months ago, almost everyone would have dismissed this.
The response has been far more interested than reflexive dismissal. Supporters see a path to people who will never open a chatbot app; others see either a huge Meta hit or a Metaverse-style miss. The difference from the Humane AI Pin and Rabbit R1 era is that the AI may finally be good enough to support the form factor.

GrokBot in Teslas
Your whole agent, now in the car.
SpaceXAI is pushing its agent onto new surfaces too, putting GrokBot inside Tesla's existing Grok voice interface, so anything managed with GrokBot can be managed from the car. Demos ranged from ordering Starbucks during Full Self-Driving to shopping for AirPods, up to running an army of agents while the car does the driving.

The Skeptics Remain
"You just connect it and it starts working for you."
Riley Brown cautioned that App Store numbers are hard to trust, comparing Muse to Threads, where users were effectively funneled in through Instagram. But at a New York poker table full of people outside the tech bubble, nearly everyone had heard of Muse and most used it, not caring about the model, just liking that it reminded them of things and canceled subscriptions.

Two Questions to Watch
Physical presence, and agents that act without being asked.
First: how much the bridge into the physical world matters for adoption. Second: whether the secret sauce is the shift from reactive to proactive agents that bring the value to users without them having to ask.