# AI Agents Are Moving Into the Real World
*The AI Daily Brief — Thursday, 2026-09-24 · https://aidailybrief.ai/e/2026-09-24*

**Personal agents are leaving the screen — and the 'nobody wants this' debate is finally meeting evidence.**

The argument used to be theoretical: Silicon Valley says consumers want the schlep work automated; skeptics like Ben Thompson say people want to shop, complain, and be entertained — not be productive. Now the evidence is arriving. Muse sits at #1 on the App Store and blew away a New York Times reviewer, Meta is pushing it onto Ray-Bans and a new always-on Charm device, and GrokBot lives inside Teslas. The people actually trying these things keep coming away more impressed than they expected.

---

## By the numbers
- **1,000** — Claude agents that ran Anthropic's enzyme hunt over 21 hours
- **210M** — Tokens burned on the discovery — roughly $10K of compute
- **5–10 yrs** — Amodei's 'cure most diseases' window if AI hits every pipeline stage
- **20** — Nations backing an international AI institution — minus the US and China
- **#1** — Muse's App Store position as the episode was recorded
- **2B** — Mothers Jill Gunter says the 'nobody wants productivity' take forgets
- **$45** — The savings Muse turned into ceramics bowls by bargaining on Marketplace
- **10 yrs** — Ben Thompson's personal assistant tenure — still not booking his flights

## Headlines

### Claude finds a CRISPR-like enzyme system `[01:00]`
Less than a week after confirming its Bay Area wet lab, Anthropic says 1,000 Claude agents working 21 hours on roughly 210 million tokens — about $10,000 of compute — discovered a novel enzyme system with a repeated-genome array reminiscent of CRISPR, with only high-level direction from scientists. Anthropic admits it doesn't yet know the system's function or significance.
Link: https://aidailybrief.ai/e/2026-09-24#claude-finds-crispr-like-system

### We've repeatedly seen a pattern where AI performance in new intellectual domains goes from weak to superhuman in a matter of a few years. `[02:00]`
*— Dario Amodei, promoting the Claude discovery on X*
Amodei's math timeline: high-school-level in 2023, competition-level in 2024, minor open problems in 2025, and the top open problems in mathematics by late 2026. He believes AI for biology is on the same exponential — the wet lab exists to physically confirm results, and the goal remains 'cure most diseases in five to ten years.'
Link: https://aidailybrief.ai/e/2026-09-24#amodei-weak-to-superhuman

### Scientists' verdict: real, fast — and very, very preliminary `[03:00]`
Genome-mining veterans note that finding a weird cluster of genes and repeats is the easy part; Anthropic hasn't shown what the system does or whether the CRISPR comparison is justified, and some see IPO-timed hype in the framing. Still, this kind of search used to take human experts months of manual work — as Tanishq Abraham put it, very cool and very, very preliminary.
Link: https://aidailybrief.ai/e/2026-09-24#cool-but-very-preliminary

### Trump rebrands AI as 'superintelligence' at the UN `[05:00]`
'The use of the word artificial makes it sound fake, and it is not fake' — the US government will now officially say superintelligence, or SI. The memes were infinite, and AI safety writers lamented a giant linguistic namespace clash with the existing meaning of 'superintelligence.'
Link: https://aidailybrief.ai/e/2026-09-24#trump-rebrands-superintelligence

### 20 nations call for an international AI body — without the US or China `[06:00]`
A Finland-led declaration backed by France, Germany, and Singapore calls for mandatory independent safety testing, common standards, and an international institution that can set standards, enable verification, and convene states when capability thresholds are crossed. Guterres is supportive; neither frontier power is on board — the glaring hole in the proposal.
*For: Legal*
Link: https://aidailybrief.ai/e/2026-09-24#twenty-nation-declaration

### A prosperous future will not be secured by a global regulator. `[08:00]`
*— Michael Kratsios, White House science advisor, to the UN Security Council*
The US position, delivered to the UN Security Council: superintelligence will be encouraged, not reined in, watched through the Department of Justice rather than constrained by new global governance structures — secured by sovereign nations, responsible companies, and 'free people who refuse to be ruled by fear.'
Link: https://aidailybrief.ai/e/2026-09-24#kratsios-no-global-regulator

### It doesn't matter whether people put their risk of catastrophe at ten percent or one percent. None of these levels are remotely acceptable. `[09:00]`
*— Sam Altman, to the UN Security Council*
Altman's Security Council speech centered on two risks: RSI producing frontier models labs can no longer understand or control, and concentration of power — 'a company or country that believes only it can be trusted with this technology can use that belief to justify almost anything else.' His middle path: neither blind optimism nor doomerism.
Link: https://aidailybrief.ai/e/2026-09-24#altman-no-acceptable-risk-level

### Amodei hands the Security Council a step-by-step governance playbook `[09:00]`
Calling AI 'the most important global security issue facing the world today,' Amodei proposed simple broad-consensus treaties like a ban on AI bioweapon development, global evaluation and verification systems so states can check rival labs' commitments, global testing standards for loss-of-control and misuse, and a global notification system. The poles are set: Anthropic on one end, the White House on the other, OpenAI in the middle.
*For: Legal*
Link: https://aidailybrief.ai/e/2026-09-24#amodei-governance-playbook

## Main episode

### People love shopping. They don't wanna be productive. They wanna be entertained. `[14:00]`
*— Ben Thompson (Stratechery), on TVPN*
Thompson's viral case against personal agents: he's had a human assistant for ten years who doesn't come within an inch of booking his flights, and consumers actively enjoy complaining about their chores — 'not only do I wanna do it, I wanna complain about it.' This is the baseline skepticism the entire category has to answer.
*For: Product, Marketing*
Link: https://aidailybrief.ai/e/2026-09-24#ben-thompson-shopping-not-productivity

### The romantic case against personal agents: people want the errand `[15:00]`
Stay Sassy argues agents don't abstract decisions, they accelerate them — ten decisions a day instead of one is 'nah, dog, a nightmare' for everyone but type-A strivers. Kurt Vonnegut made the same point in 2005 about walking out to buy a single envelope: 'we're dancing animals... we love to move around.'
Link: https://aidailybrief.ai/e/2026-09-24#romantic-case-against-agents

### The counterpoint: nobody loves schlep work `[16:00]`
Amy Wu Martin: there are window shoppers and there are affluent time-buyers who already hire stylists and travel agents. Simon Taylor: booking a kid's cycle class across forms, calendars, and deposits is 'buying things, not shopping' — delegation gold. Jill Gunter: if you think nobody wants to outsource admin, you're forgetting the world's two billion mothers. Madhu Guru: consumers don't have one relationship with doing things.
*For: Product*
Link: https://aidailybrief.ai/e/2026-09-24#schlep-work-counterpoint

### Muse is #1 on the App Store — and the buzz is coming from unexpected places `[18:00]`
Yes, a company with Meta's distribution can drive an app to the top of the charts even without resonance. But the more telling signal is genuinely positive reviews arriving from well outside the usual fan base.
Link: https://aidailybrief.ai/e/2026-09-24#muse-tops-the-app-store

### The NYT handed Muse its life — and was blown away `[19:00]`
Eli Tan's agent filled out online forms, paid a photographed parking ticket with approval, worked his dental insurance's customer service by phone using different personalities, and turned $45 of savings into Heath Ceramics bowls by messaging and bargaining with Facebook Marketplace sellers. His verdict: 'the most useful AI app I had ever used.'
*For: Product, CS*
Link: https://aidailybrief.ai/e/2026-09-24#nyt-blown-away-by-muse

### The skepticism just moved from utility to privacy `[20:00]`
The most interesting part of the NYT experience: the reviewer came in doubting whether personal agents were useful enough, and came out doubting whether enough people will be comfortable handing over their most personal data to unlock that usefulness. That's a very different bottleneck.
*For: Product, Legal*
Link: https://aidailybrief.ai/e/2026-09-24#skepticism-shifts-to-privacy

### Meta is grabbing this momentum by the throat `[21:00]`
Chief AI officer Alexander Wang is on a partnership frenzy — Spotify, Box, Dick's, Sephora, Gap, Walmart, and more — while Muse ships computer use on Mac and a real-time video avatar. Watching the Meta team on social, they clearly believe they have something.
*For: Marketing, Product*
Link: https://aidailybrief.ai/e/2026-09-24#meta-grabs-momentum

### Meta Connect was built around getting Muse into the real world `[22:00]`
After years of anticlimactic Connects, this one had no such problem: a Ray-Ban line refresh including a first-ever audio-only, camera-free pair, with hands-free Muse control listed first in Zuckerberg's rundown — long conversations while 'Muse will work in the background and get things done for you as you're talking.'
*For: Product*
Link: https://aidailybrief.ai/e/2026-09-24#meta-connect-built-around-muse

### One more thing: Muse Charm, an always-on agent device `[22:00]`
A new hardware category about the size of an AirPods case, with a small color screen and direct voice instructions — no phone unlock, no app. Zuckerberg calls it 'by far the fastest way to talk to your Muse' for non-Ray-Ban owners; it's still in active development, but Meta hopes to ship by the holidays.
*For: Product*
Link: https://aidailybrief.ai/e/2026-09-24#muse-charm-one-more-thing

### Why this isn't the Humane Pin: the AI might finally be good enough `[24:00]`
Silicon Data's Steve Howe notes the Humane AI Pin and Rabbit r1 flopped because the AI couldn't support the bold form factor — 'I think this may finally have changed.' Six months ago almost everyone would have reflexively dismissed a device like Charm; the initial response this time is genuine interest.
*For: Product*
Link: https://aidailybrief.ai/e/2026-09-24#why-this-isnt-the-humane-pin

### GrokBot moves into the Tesla dashboard `[24:00]`
SpaceX AI launched GrokBot in Teslas with no separate app — it lives inside the existing Grok voice interface. Owners are ordering Starbucks while on full self-driving, shopping for AirPods, and managing email, calendar, and finances by voice; one builder described having the car drive while he talks to 'an army of agents.'
*For: Product*
Link: https://aidailybrief.ai/e/2026-09-24#grokbot-moves-into-teslas

### The poker-table read: doomers who use Muse daily `[26:00]`
Riley Brown's non-tech NYC poker game — bankers, a doctor, insurance — shared the assumption that AI leads to human extinction within ten years, and most were Muse users anyway. Their description of why they like it: 'you don't have to ask it to do stuff, you just connect it and it starts working for you.' They didn't care about the model at all.
Link: https://aidailybrief.ai/e/2026-09-24#poker-table-read-on-ai

### Two questions to watch as personal agents chase the mainstream `[27:00]`
One: how much does the bridge into the physical world — glasses, cars, Charm — actually matter for adoption? Two: is the secret sauce the shift from reactive to proactive agents that bring value without being asked? For now, the winds are blowing in agents' favor: people trying these things keep coming away more impressed than they expected.
*For: Exec, Product*
Link: https://aidailybrief.ai/e/2026-09-24#two-questions-for-personal-agents

*Today's sponsors: KPMG, Blitzy, Robots and Pencils, Hyperagent — offers at https://aidailybrief.ai/sponsors*

---
Transcript: https://aidailybrief.ai/e/2026-09-24/transcript.md
Listen: https://pod.link/1680633614 · Ad-free: https://patreon.com/aidailybrief
© 2026 The AI Daily Brief — Until next time, peace ✌