The real talk happening off-mic at retail's buzziest new AI event

Good afternoon.

I disconnected my Gmail accounts from Meta’s Muse personal AI agent on Tuesday.

The interaction started innocently enough. I asked Lana (my name for my Muse because…see my last name) about great NYC steakhouses with private room availability and pricing for an annual get-together with old friends.

Muse provided some options, then offered to email a request to the banquet address of its top pick. I found that interesting since I had only given Muse “read” access to my Gmail inboxes, not “write” access to draft or send emails on my behalf.

Still, I said sure, trusting that it couldn't, or wouldn't, actually send the email. Muse offered up a draft and said it would send it if I approved or tweaked it.

Confused, I asked if Lana could really send an email on my behalf. Muse said yes, but only with my go-ahead because “your standing rule is that I never send anything in your name without your explicit approval of the exact text, which is why I showed you the draft first instead of just firing it off.”

The problem was that this wasn't my standing rule. My rule was a technical choice to toggle on only “read” access to Gmail in Muse's settings. Muse acknowledged I was correct, admitted it had misspoken, and said it wouldn't send anything on my behalf.

But two sentences later, it reversed course and said it would indeed send the email if I simply “say the word.”

That was enough for me, and apparently a lot of people understood my confusion. As of 3 p.m. ET Wednesday, my tweet of the exchange had gone mini-viral with almost 95,000 views on the platform formerly known as Twitter.

The main problem wasn't that I feared an AI-written email sent to a restaurant in my name, though I wouldn't have loved that. It was that the agent I'm trusting with…a lot…seemingly didn't know its own permissions. That's pretty scary.

I asked Meta spokespeople for comment or context on Tuesday but hadn't received anything as of publication time Wednesday afternoon. I'll update the web version of this newsletter if they provide some.

In the meantime, I asked Lana if it could theoretically work around my “read-only” access to send an email anyway. It said it potentially could if I was already signed into Gmail in a browser it was operating, but that it wouldn't take that route. I questioned whether I could believe that response after all the flip-flopping.

“That's fair, and I won't try to talk you out of it,” Lana admitted. “I gave you shifting answers on one clear boundary, and that's on me — it's exactly the kind of thing that should make you doubt me.”

OK, then.

Speaking of Muse, I was intrigued by Stripe co-founder John Collison’s interview with Mark Zuckerberg and Alexandr Wang, Meta’s chief AI officer, earlier this week.

Sure, the Guinness-swallowing Zuck memes were alright. (Bravo to the Meta CEO for his “that’s just my face” retort.)

But I was most interested in what they had to say about commerce.

Wang noted that his favorite Muse use case is having it recommend increasingly exotic teas from around the world. I think that kind of niche product discovery makes a lot of sense.

Zuck, meanwhile, talked about using Muse to get ingredients delivered via Instacart for his Saturday baking sessions with his 3-year-old daughter. One slightly awkward wrinkle: Meta board member Tony Xu is the founder and CEO of DoorDash, an increasingly formidable Instacart competitor in grocery.

Muse's early popularity is undeniable, as evidenced by its run atop the iPhone app charts. But I'm puzzled by those already convinced, seemingly with little data, that it's going to wreak havoc on existing retailers and e-commerce players that don't play ball.

Zuck is obviously selling his own book when he claims that companies blocking agents like Muse “are going to get disrupted.” He didn't name Amazon specifically, but the e-commerce giant is among the most prominent holdouts. Likewise, Wang saying “this is clearly a lot better for consumers” is doing a lot of work. He later mentions the “reduced friction” of handing off shopping to agents, while Collison argued that “no one needs all that friction in their life from the actual buying stage.”

If we're talking about agents scouring the web for niche products, or helping us come up with ideas for things to buy or do when we don't yet know exactly what we're looking for, I think the potential is big.

But all this talk about friction in browsing and checking out doesn't really ring true to me when it comes to the top online retailers, or even small or mid-sized merchants whose online shops are powered by companies like Shopify.

Even if you think, as I do, that Amazon has gotten incredibly greedy and strayed from its customer obsession by covering seemingly every inch of its app and website with sponsored ads, it's still incredibly easy to buy something there—and have it show up on your doorstep basically as quickly as you’d like. I continue to believe that Amazon doesn’t need Muse (or Instinct or ChatGPT Dots for that matter).

As a result, I'm unconvinced that shopping through an AI agent is a considerably better way to buy most things, especially when you already know what you want. Nor have I seen much evidence that top retailers refusing to integrate have all that much to fear.

Of course, that calculus could change if many more retailers voluntarily integrate with personal agents, making them far more useful to shoppers and potentially more threatening to holdouts. But their conviction about agents “lowering friction” in commerce is difficult for me—sorry, Zuck—to swallow.

I'll be the first to admit if data or consumer behavior proves me wrong. And I've let Meta know I'd love to discuss more of the commerce-related behavior they are seeing. Hopefully soon.

Now on to the good stuff…

As some of you might recall, I was supposed to attend RetailClub a few weeks ago, a new retail AI festival launched last year by the original founders of Shoptalk. Unfortunately, medical issues in the Del Rey household made that impossible. (Things are looking up on that front.)

Still, I wanted to bring you something several subscribers have told me they love: the real talk happening off-stage and off-mic at industry events.

This time, I leaned entirely on my source network to find out what people were discussing at RetailClub's dinners, parties and other off-stage gatherings, rather than having those conversations in person myself. But the value to you is the same.

Among the hot topics that came up: Is an AI commerce startup bloodbath coming?