Meta launched a personal AI agent called Muse on Tuesday. Asked inside Meta's own assistant how to use it, the answer that came back was that Muse is a free AI assistant from Anthropic.
The screenshot came from Opendoor's chief executive, who had posted it while asking Meta to put the agent inside WhatsApp. The hosts could not find anyone else discussing it, and there is no product called Muse at Anthropic for the model to have confused it with.
"So, I'm looking at this post. I'm trying to find why is nobody talking about how Meta AI thinks that Muse is a product from Anthropic and is telling people to go log in there."
John Coogan and Jordi Hays present TBPN live every weekday and worked the day's news for the first hour before the first of six guests arrived.
I listened to the full segment so you can skip it. 52 minutes of audio, 17 minutes of reading.
Here are the 10 takeaways that matter.
🎙️ Hosts: John Coogan and Jordi Hays, who present TBPN live on X and YouTube every weekday from the studio they call the Ultradome
📰 Published: 9 September 2026 on YouTube (TBPN)
🔴 YouTube | 🟣 Apple Podcasts | 🔗 Show notes | ⏱️ 52 min | ✅ Time saved: 35 min
Key Takeaways
Apple's first foldable, the iPhone Duo, appears to have solved the crease
The hosts expect two to three thousand dollars fully specified, and expect it to sell
Their verdict on the whole event was that it changes nothing about what a phone does
Meta's Muse is free at the bottom with $20 and $100 monthly tiers above it
It is against capital spending the Journal reports will top $130 billion this year
Meta's own assistant answered a question about Muse by describing it as an Anthropic product
Meta wins categories by putting them inside Instagram, not by launching separate apps
Muse is a separate app, which the hosts think makes it harder
An Anthropic researcher resigned publicly, saying both his employers are gambling with our lives
An Anthropic alignment researcher replied agreeing, and put extinction above 10% within a decade
A 1985 Steve Jobs clip describes a conversational interface to a thinker's worldview
The hosts think self-regulation among the labs runs into antitrust before it runs into anything else
1. The iPhone Duo Lands
The product is Apple's first folding phone. "Well, Apple has announced the iPhone Duo, their foldable phone."
The thing they went looking for first was the fold line. "So in this video, there is no crease." Their read is that Apple waited long enough for the technology to get past it, and one host said he would be shocked to see a visible crease
They expect it to sell. "I think this is going to sell like hot cakes. I think people are ready to spend." Their price expectation: "It's going to be like two grand, maybe three grand when you max it out."
The buyer they described is one who wants the purchase to show. A Pro model costs 30% more; this one signals something larger than an incremental upgrade
Their economic argument for premium phones is time. "phones have always been incredibly cheap" relative to the value people get from them, given that most people spend more time on a phone than in a car costing ten times as much
The verdict on the product itself was flat. "This feels like the biggest step forward in a long time and yet it does absolutely nothing new for you." One host called that a rave review and added that it does not move humanity forward at all
The genuine question they raised is what it does to app design. They used the guest they had coming later, Suno, as the example: a full-featured phone app, but the composer that deconstructs a song only works on a desktop. Their expectation is user-interface opportunities and challenges in equal measure
The running joke was screen time. If two screens are open, does screen time run twice as fast, and should it be adjusted for pixels — an hour on a watch counting as fifteen minutes of a triple-monitor setup
2. Apple Down 1.27% On It
The market reaction was mild. One host read the tape out as Apple "down 1.27% today" and "down 5% over the last five days" while sitting flat on the month, and said all of it had been priced in
The event mattered more for the new chief executive than for the product. The hosts read John Ternus's first major event as a strong opening: a genuinely new product rather than a minor refresh
They also read it as a managed risk. Putting him on stage with a foldable rather than something like the Vision Pro means his first outing is not remembered as a flop
They were sympathetic about what users actually want instead. Offered a choice between the foldable and working search inside iMessage, one host took search. A post they read out captured the other half of the mood: "Mentally ready for my iPhone to start being slow, laggy, and having a bad battery."
The rest of the line got the standard treatment. The iPhone 18 keeps the same shape and is thinner, lighter and faster; Apple Intelligence was updated; and the hosts believed this was the launch at which the new Siri actually ships
One host disclosed that he had seriously considered carrying an iPad mini as his daily phone, because he dislikes phone calls, and had worked out that two-factor codes could be routed elsewhere
3. Live Rewind on the Watch
The hosts played the Wall Street Journal's Joanna Stern demonstrating a new Apple Watch feature.
What it does. "Just double press the digital crown in live rewind transcribes what was just said." The demonstration was a waiter reading out the specials, with Siri able to answer questions about the transcript later
Her objection was the one the hosts repeated. On the restaurant example: "maybe just ask the waitress to repeat herself."
The hosts found a case where the objection fails. A lawyer walking through a contract and flagging the parts that matter — you could ask them to repeat it, except "it costs a thousand bucks an hour"
Their broader complaint was about demonstrations, not features. They contrasted it with the GarageBand demonstrations of the 2000s, where the alternative to the software was obvious and the appeal was immediate, and said these launches get harder to make feel natural
On the Apple Watch Ultra 4 with the chunky metal bracelet, the verdict was short. "It's horrific." The complaint was the clasp, which bolts on at two points rather than integrating with the case. They conceded a heavy strap may suit a watch that has itself become heavy, and then suggested launching it as an ankle bracelet
4. Daring Fireball, 354 Days
John Gruber reviewed last year's phones on the eve of their replacement. The hosts read his own explanation: an unplanned personal hiatus meant he missed the review window, and "354 days is arguably more than a few"
His verdict on the model being superseded was the strongest line in the review. "the iPhone Air is the nicest phone Apple has ever made" — probably the nicest device the company has made in any category
One host said he is waiting on the same writer's review of the foldable before forming a view
5. Jobs Described ChatGPT
The hosts played a 1985 clip of Steve Jobs speaking at a university, which one of them had found after a viral post.
The setup is Aristotle. Jobs asks who tutored Alexander the Great for fourteen years, says he became jealous reading it, and notes that the printing press lets him read what Aristotle wrote without an intermediary
The limit he names is the one large language models removed. He cannot ask Aristotle a question and get an answer, and his hope is a tool of a new and interactive kind that captures a thinker's underlying worldview so a student can ask it questions
The hosts' reaction was to the specificity. One called it compressing the data down and putting a conversational interface on it, and said "It's a crazy prediction 40 40ome years ago."
They spent as long on the delivery as the content. Jobs lands the joke about not getting an answer and lets the laugh sit, which they compared to the timing of a well-known technology reporter
They then asked an AI version of Jobs what he would think of AI today. The answer they read out set three conditions: give people new abilities, make it effortless enough to work from minimal instruction, and make something excellent rather than large quantities of generic output. The conclusion was that he would be against low-quality mass output and in favor of AI
The imagined Aristotle answer went further. It cared more about what AI does to character than about how intelligent it becomes, and favored AI as an instrument of human flourishing while resisting the idea that a life requiring less effort is automatically better
The hosts' own question was whether that vision could survive today's discourse. One argued that political polarization, economic stagnation, inequality and negativity bias on social media make the urge to tear a figure down unavoidable, so there would be viral videos attacking the keynote regardless
The other host's answer was that the technology itself is frightening. He said he has come round to that view, that some actors are positioning it as more frightening than it is, and that even with Jobs alive it would still be a mess. Asked what Jobs's probability of human extinction would be, the answer offered was 1%
The digression they enjoyed most was Jobs introducing GarageBand with John Mayer playing guitar beside him on a stool
6. Meta Launches Muse
The hosts noted where the story ran. A proof of the Navier-Stokes equations did not make the print edition of the Wall Street Journal that day; Meta's agent did. One host's explanation was commercial: it is hard to make money selling mathematical theories
The product, in the Journal's description they read out. "The agent called Muse is accessible via dedicated app," operated by conversation like a chatbot, with a user-chosen name, able to buy goods online, respond to emails and complete other authorized tasks. The hosts believe there is a Stripe payments integration behind the purchasing
The financial framing was the capital spending it has to justify. Meta's AI investments are, in the paper's words, "expected to top 130 billion this year"
The pricing. Free for most of what people need, with two subscription tiers for heavy users at "100 bucks a month and $20 a month"
The technical detail the hosts found most interesting came from Ben Thompson. Each user gets a virtual machine — "eight gigs of memory and 8 gigs of storage that is yours sequestered" — with end-to-end encryption of the kind Meta built for WhatsApp, and the hosts noted the involvement of Moxie Marlinspike, who created Signal
Their doubt is the unit economics. A machine that spins up on use and shuts down after still costs real compute hours, and $20 or $100 a month is roughly what an ordinary virtual server costs to rent outright
The competitive read from the Journal. Anthropic and OpenAI have built multi-billion-dollar businesses on agents, but most of that usage is coding and business applications; consumer adoption lags, which leaves room for Meta and Apple
Meta's own pitch, from its AI chief. Alex Wang told the paper that packaging a powerful technology so that it is genuinely digestible is what will set it apart from other agent products
7. Muse's Anthropic Answer
The screenshot came from Opendoor's chief executive, posted alongside a request that Meta make WhatsApp the interface for these products
He asked Meta's assistant a simple question and got the wrong company. "Muse is a free AI assistant from Anthropic. You can use it on the web, on mobile, or through their API." The follow-up instruction was to go to Anthropic's site, create an account and log in
The hosts could not find anyone else on it. "So, I'm looking at this post. I'm trying to find why is nobody talking about how Meta AI thinks that Muse is a product from Anthropic and is telling people to go log in there."
What made it strange to them was the absence of any link between the two. As far as they knew, Anthropic has never had a product called Muse
Their explanation is version drift inside Meta's own estate. Meta AI runs an older model, and a user who ends up in a chat inside Instagram lands on a different model again. Unification is needed and is hard, because deploying a new model at the scale of billions of users is not a day-one operation — the comparison they drew was that Google cannot simply drop its newest fast model into search on launch day
8. Distribution Beats Novelty
The hosts' historical case is Instagram Stories. Meta did not beat Snapchat by shipping a separate clone; it won by putting the feature inside an app that already had the installed base
Which is why a standalone app worries them. If Muse works its way into Instagram it probably gets adoption; as a separate download it is trickier
They allowed the counter-example. Threads was routed out of Instagram, appears there, carries advertising and reached what they described as hundreds of millions of monthly users — though they discounted that number for people who arrived through clickbait, while accepting there is a genuine community of Threads users
The asset they think is decisive is free inventory. Meta can build advertising slots for itself at whatever specification it likes and put a billboard in front of everyone who opens any of its apps — something they expect it to hold back until the early adopters have found the problems
The early ranking was strong. "Muse is number three on the charts in the app store" behind a fantasy sports app and ChatGPT
On whether it is a copy of a rival product called Instinct, the hosts said no. Their evidence: reporting that Meta has worked on it for almost a year, the Manus acquisition, and Mark Zuckerberg's public writing about personal superintelligence. They also recalled interviewing him a year ago and being told he was not ready to announce shopping inside the assistant, while it was clear that was the direction
They still expect a fight. Instinct is early, and it faces a company with resources going hard at the same problem. The hosts' complication is that people come to Meta to be entertained and Muse looks like a productivity product — though they noted how much shopping already happens through Instagram browsing and retargeting
One post they read out was a screenshot of Zuckerberg's early Facebook messages, quoted back at the launch announcement
9. Zuckerberg's Irish Castle
The first item was his face. The hosts compared photographs and concluded his nose is starting to resemble that of the UFC champion he has been training with, which they read as evidence he is training harder than he admits
The second was the property. They played a video breaking down a Gothic revival castle he has bought on the banks of the river Blackwater in County Waterford, designed for a local landlord who was a member of parliament and high sheriff of the county. "The house was complete in 1830," and the grounds hold a folly built in 1826 and a former aviary
The detail that landed was the walk. According to the records read out in the video, it took "a brisk 4 and 1/2 minute walk" to get from the kitchen to the dining room
The hosts' serious point was about carrying costs. Some castles are cheap relative to how impressive they are, and expensive to hold: cleaning alone runs to six figures a year before landscaping
Their answer to where technology wealth ends up is patience. Either the owner tires of it, donates it or passes it on, or it leaves the family several generations later — and what is left is a museum brought up to modern standards
They noted a run of technology chief executives appearing on the same comedy podcast, and said they were disappointed by which ones had not
10. An Anthropic Resignation
A researcher resigned publicly and quit the industry, not just the company. The post read: "I resigned from Anthropic today. I spent the last three years doing pre-training research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving super intelligence and gambling with our lives."
The reach was enormous. The hosts put the post at "a 100 million views and 600,000 likes." One of them predicted it ends up on a political billboard
An Anthropic alignment researcher agreed on the record. Evan Hubinger said the resigning researcher is correct: "We really do earnestly believe AI could kill all humans. I personally think it's above 10% within the next decade." He added that he believes Anthropic is trying its best, that there is no plan yet to solve alignment for superintelligence, and that the company is clearly not on track to
The pushback the hosts read out came from several directions. Martin Casado of Andreessen Horowitz said he worked on an actual thermonuclear weapons project and found the dissonance less bizarre. Nick Carter argued that anyone who genuinely believed this would be sabotaging their employer rather than posting. Others asked whether the departing researcher was giving back his equity
The demand that recurred was for a mechanism and a falsification test. One post the hosts quoted approvingly asked anyone assigning a greater than 10% probability to human extinction within a decade to explain how they reached the number and what would change their mind. Their own read was that many people found the estimate under-researched, though Hubinger had linked to Anthropic's published risk report
Eliezer Yudkowsky's response was to thank the researcher for saying plainly what he believes. The hosts noted that his co-author Nate Soares was booked on their show the following day to discuss their book, "If Anyone Builds It, Everyone Dies"
One reading of the whole thing was bullish. Ryan Petersen's post, read out on air: "I need to buy more stock in this company. I've never seen such a commitment to making money regardless of the consequences."
The hosts' own substantive worry is legal. If the labs respond to the absence of regulation by coordinating among themselves — pacing the frontier, collaborating on safety — that runs into competition law: "you're not allowed to price fix, you're not allowed to collude"
Their second observation is about the credibility trap around lab predictions. When leaders forecast, outsiders say they are talking their book; when they ask for regulation, outsiders call it regulatory capture. What they say they are now seeing is people who criticized the labs deciding they want regulation anyway and accepting the capture that may come with it
The reason they gave for the mismatch is information timing. People inside the labs see the next capability step earlier, so their predictions arrive before anyone outside can digest them
One host also argued that resigning is not obviously the most effective move, because in a large organization a person at a given level often cannot get a decision made, however many messages they send
Bonus Insights
The show opened on a countdown to Christmas — 106 days — and a suggestion to remove the floorboards and install a permanent tree, which is where the running joke about a house built around a Christmas tree came from
A parody resignation post circulated the same morning, in which someone claimed to be quitting television advertising because his employers were racing toward self-improving advertising performance and gambling with the audience's return on investment. One host read only the first line before realizing it was a joke
The hosts flagged that they had missed the day's other big story, the theft of paintings from a French museum, and promised to return to it. They did not
On Meta's rumored interest in the rival product, the post they read said a ten-figure offer had been made for a company roughly a year old and turned down
The hosts' bottom line is that Apple shipped the most impressive hardware in years without changing what a phone does, while Meta shipped a consumer AI agent whose own parent assistant does not know it exists — and that the loudest argument in technology this week was not about either product but about whether the people building the models believe they are safe.
Products, Companies & Tools Mentioned
Apple (Launched the iPhone Duo, its first foldable, plus the iPhone 18 and an Apple Watch Ultra the hosts disliked; the stock fell 1.27% on the day)
Meta (Launched Muse, a personal AI agent that shops and answers email, against AI capital spending the Journal reports will top $130 billion this year)
WhatsApp and Instagram (The distribution surfaces the hosts think decide whether Muse works — Meta won Stories by embedding, not by launching an app)
Anthropic (Named wrongly by Meta's own assistant as Muse's maker, and the employer at the center of the day's resignation)
OpenAI (Named alongside Anthropic in the resignation post, and cited by the Journal as one of the two companies with real agent revenue)
Stripe (The payments integration the hosts believe sits behind Muse's ability to buy things)
Signal (Referenced through Moxie Marlinspike's involvement in Meta's encryption work on the new agent)
Opendoor (Its chief executive posted the screenshot in which Meta AI attributed Muse to Anthropic)
GarageBand (The hosts' example of a product demonstration that sold itself, against the new Apple Watch feature that did not)
Books & Resources Mentioned
Steve Jobs Predicts ChatGPT in 1985 (The clip the hosts played in full — Jobs on wanting to ask Aristotle a question and get an answer)
Meta Launches a Personal AI Agent Designed to Be Easy to Use (The Wall Street Journal piece the hosts read from on Muse, its pricing and its capital-spending backdrop)
Anthropic Researcher Quits Over 'Out-of-Control' AI Fears (The Journal's report on the resignation that dominated the second half of the news block)
Apple to Debut New Foldable iPhone, New CEO and New Prices (The Journal's preview of the event)
The iPhone Air and iPhone 17 Pro (John Gruber's review, published 354 days late, calling the iPhone Air the nicest phone Apple has ever made)
Here's how AI could kill us all (if the worst fears come true) (Axios on the extinction-probability argument the hosts spent the last section on)
If Anyone Builds It, Everyone Dies – Eliezer Yudkowsky and Nate Soares (The book behind the following day's booking, which the hosts trailed during the doom discussion)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: