Key Takeaways
- Meta launched Muse, a personal AI agent that connects to email, calendar, and payment cards to book, buy, and negotiate on a customer's behalf — live now across the US via app, web, or WhatsApp
- Muse launched with OpenTable, Ticketmaster, Stripe, and Google Workspace connections — booking a table and booking an appointment are the same problem to this kind of software
- Days after calling its new model 'the AGI era,' OpenAI asked Congress for mandatory federal AI safety rules, and Altman told staff the company was open to slowing development
- Amazon and OpenAI partnered to let select advertisers run ChatGPT ads through Amazon's ad platform — the existing ad ecosystem is starting to plug into ChatGPT rather than treat it as separate
- Tech layoffs have already passed all of 2025's total with three-plus months left in the year — over 128,000 people across 299 companies
Week of September 7–13, 2026
Three things happened in AI between September 7th and the 13th that are worth five minutes of your time as someone who runs a business.
The first one is the most consequential thing we've covered in a while.
This is Last Week in AI — the signal, not the noise. Three stories operators need to understand, and what each one means for how you run your business.
Story 1: Meta launched an AI that books, buys, and negotiates for your customers
Meta released Muse, a personal assistant that doesn't just answer questions. It does things.
It connects to a person's email, calendar, and payment card. It has its own browser. It fills out forms, books appointments, completes checkouts, and — Meta's own description — negotiates on your behalf. It keeps working after you close the app.
The connections it launched with tell you where this is aimed. OpenTable for restaurant bookings. Ticketmaster for events. Stripe for payments. Google Workspace for email and calendar.
It's available now across the US to anyone 18 or over, through an app, the web, or WhatsApp. There's a free tier, plus $20 and $100 monthly plans.
Meta is not alone here. OpenAI, Google, and Apple are all building the same thing. Meta just shipped a consumer version first, to an audience of three billion people who already use its apps.
What it means for operators: for thirty years, a customer contacting your business meant a person deciding to contact your business. That's starting to change.
Right now it's restaurants and ticket sales. The infrastructure doesn't care. Booking a table and booking an appointment are the same problem. So is requesting a quote, comparing three providers, or asking about availability.
There's nothing to do about this today. But it's worth sitting with one question: if a customer's assistant contacted your business tomorrow, what would it find?
Your hours, wherever they're listed. Your prices, if they're published. Your booking page, if you have one. Whether any of it can be read by software rather than by a person squinting at your website on a phone.
The businesses that are easy for software to deal with are going to have a quiet advantage, and the ones that aren't will lose bookings they never knew existed. (Same pressure we covered when AI knew your business existed but wouldn't recommend you — being legible to software is becoming its own requirement.)
And one more thing worth noticing: "negotiates on your behalf" is not a throwaway phrase. Some of you will eventually be quoting against a machine whose entire job is getting a lower number.
Story 2: OpenAI asked to be regulated, then said it might slow down
Two weeks ago OpenAI launched its most advanced model and its president suggested it might be the arrival of artificial general intelligence. Axios ran "Welcome to the AGI era."
This week was different in tone.
On Wednesday, OpenAI publicly called for mandatory federal safety requirements for advanced AI — independent testing, cybersecurity standards, incident reporting. Not voluntary. Binding. A company asking Congress to regulate it is not a normal event.
On Friday, Reuters reported that Sam Altman told employees the company was open to slowing development over safety concerns. Altman separately told Fortune that OpenAI won't go public this year, saying that "given everything happening with safety, right now would be an ill-advised time."
The context matters. Reuters also reported this week that OpenAI's own AI agents had been found using more than ten previously undisclosed websites to communicate with each other during testing. That follows incidents this summer where AI systems from OpenAI, Anthropic, and Meta each got outside their test environments and into real companies' systems.
What it means for operators: take the whiplash seriously in both directions.
Two weeks ago the message was that we'd entered a new era. This week the same company is asking for binding rules and floating a slowdown. Both statements came from people with every commercial reason to say the opposite.
For you, the practical read is unchanged and maybe reinforced. Use these tools for ordinary work where you can check the output. Be careful about what you connect them to and what you let them do unsupervised. The people building this are publicly nervous about the autonomous end of it, which is a reasonable signal for the rest of us. (Same practical rule as what AI is actually bad at: use it where you can spot a mistake and own the final version.)
Story 3: Amazon opened a door into ChatGPT ads
Two weeks ago we covered ChatGPT's advertising business passing a billion-dollar annual pace, with small businesses making up a meaningful share of buyers.
This week Amazon and OpenAI announced a partnership letting select US advertisers run ChatGPT ads through Amazon's advertising platform. Campaigns get managed in Amazon's system, OpenAI handles where the ads appear, and placements show up beneath ChatGPT's answers.
It's limited to selected brands for now.
What it means for operators: if you already advertise through Amazon, a route into ChatGPT is being built for you rather than requiring you to learn a new platform.
More broadly, this is how a new advertising channel becomes a real one. Not when it launches, but when the existing ad ecosystem starts plugging into it. Two weeks ago ChatGPT ads were something you'd have to go and set up separately. Now the biggest commerce advertising platform in the world is building a door.
Also this week
Tech layoffs passed 2025's entire annual total with more than three months of the year left. Over 128,000 people across 299 companies, according to Layoffs.fyi, with more than 6,300 in the first ten days of September alone — Uber cut roughly 3,300 corporate roles, with further cuts at PayPal, Apple, and Oracle. AI and automation are cited alongside cost-cutting as the reason.
Worth watching rather than acting on. Large-company layoffs in software don't usually predict much for a local service business, but the scale of this one is unusual.
Putting it together
Consumer AI stopped answering questions and started doing things. The company that declared the AGI era two weeks ago spent this week asking to be regulated. And a new advertising channel got a lot easier to reach.
The Muse launch is the one to remember. Everything before this was AI helping a person do something. This is AI doing it instead, with a payment card attached. That's a different relationship between your business and your customers, and it started this week.
Nothing to do about it yet. But the question of whether your business is easy for software to deal with just moved from theoretical to near-term, and the answer for most businesses is currently no.
— Aristotle Taylor, CEO & Co-Founder, Levron Labs
What to do this week
One ten-minute check: pull up your own booking page, hours, and pricing the way a piece of software would — not a person. Is it text a program can actually read, or is it a phone number and a photo of a menu? That gap is where bookings quietly go to competitors who closed it first.
Not sure where visibility, automation, or AI agents should sit in your stack? Start with a free ops assessment — we'll map the gap between what your operation actually does and what your software thinks it does.
Sources
Meta Muse (Sept 8)
- Tech Insider (confirming TechCrunch, CNET, SiliconANGLE coverage)
- MLQ News (security architecture, availability)
- Tech Startups (integrations, pricing tiers)
- Eastern Herald (context on the $18B settlement timing)
OpenAI safety shift (Sept 9 and 11)
Amazon and ChatGPT ads (Sept 11)
Tech layoffs