This week's AI news felt, for the first time, less about the fast and furious model improvements and more about what normal people might do with them.
Meta released Muse in the US and people have immediately gotten over their mistrust of Zuckerberg's personal data driven ad empire because the product seems to be just that good. Here's an early adopter anecdote from a former Salesforce colleague Mick Costigan:
He put his Muse, Dusty, to work planning his family’s meals and date nights. Every Sunday, Dusty scans the family calendar, picks recipes from NYT Cooking, builds a grocery list, and purchases everything needed at his local grocery store, King Soopers.
If I've learned anything over the last year of working on agentic AI for a grocer it is that Silicon Valley hates grocery shopping, so any Muse PM's reading that quote will be over the moon. But, for my purposes, it neatly shows how a tool being really useful can overcome trust barriers.
Muse is currently #1 on the app store and, reportedly, growing faster than ChatGPT did in it's early days with 730,000 downloads in the first five days (though it is early days). It's cute mascot must be helping. Though the further news this week was Meta is launching the Muse Charm. Already being compared to a tamagotchi, this device is in the vein of the Rabbit R1 or Humane Pin and, from my very unscientific sampling, it is driving most of the negative perception of Muse given it seems incredibly kid-oriented and Meta have been fined(ish) $18bn to resolve claims its platforms harm kids.
My assumption is that the device is a move to create an advantage against Apple and Google/Android. It's a bet on a future with an agent-first interface. That is a world some companies are starting to get ready for.
Last week, Brian Morrissey of People vs Algorithms fame (for me at least) hosted a conversation with Nick Thomson the CEO of The Atlantic. Let's remind ourselves that The Atlantic is a 169-year-old publication: they've survived a fair amount of disruption. He painted three scenarios for AI technology plays out and impacts media. (I'm cribbing these from BrXnd Dispatch).
Scenario 1: Precedented Times: AI technology is overrated. It becomes a "normal" technology and is integrated into products and services making them a bit better for customers.
Scenario 2: The AI Monoculture: AI disrupts the entire structure of technology, capital, and labor. "Most digital and technology businesses as we know them today will cease to exist."
Scenario 3: AI Intermediaries: AI becomes the primary way people interface with the digital world. Personal agents synthesise media, products – everything we look for on the internet – into digests for their user.
Related, OpenAI are trialling a new ad experience. Clicking an ad in ChatGPT takes you to a "sponsored conversation" to find out more about the product or service. It's not yet clear if that's a conversation with a specific system prompt or a RAG lookup or how much control the advertiser has over that conversation. But it's another hint at a slightly different version of the AI intermediary scenario. This ChatGPT sponsored conversation feels like a prototype for an agent to agent conversation. Today we click through to a website to find what we're looking for. In this world, we ask questions directly to get the information we need. This can be incredibly disruptive for retailers as it means the way category norms and heuristics are structured today breakdown. There are only the user needs, the product specs, and the AI intermediaries' abilities to match the two.
I've been touting the possibility of a version of scenario three for a few years now so it's really nice to hear it being talked about by the CEO of The Atlantic. The fact that they're taking steps to prepare for it, feels like a signal in itself.
At this point, the gut-check kicks in. Does anyone actually want this? And while I don't think adopting the behaviours of rich people wholesale is aspirational, what I call the 'entourage check' can be informative. When people get rich, they hire more and more people to advise them on decisions, to help them with things they don't want to take care of. The richer they are, the more the entourage grows. The entourage check is whether an AI-use case is done today by a person on behalf of someone rich. Will people let an AI agent mediate their media consumption? Well, we know some people have assistants who read the press and then tell their employer what they need to read, or give them a synthesised briefing on it. This passes the entourage check.
Muse and the Charm are evolutions of OpenClaw and the AI-first devices. But they're from a company with huge resources and massive reach and Muse has been downloaded 730,000 times in its first five days.
The signals are getting stronger.