Summary

OpenAI’s reported mobile, screenless speaker looks like a disciplined V1 rather than a premature phone replacement. Ben argues that the home is where people can comfortably talk to a device and may not have a phone with them because it is charging; strong microphones could pair that natural interface with capable AI. His hedge is explicit: “It’s possible it doesn’t work,” but a failed V1 would not require going all the way into building a phone and ecosystem.
Portability is the strategic differentiator because it can approximate ambient AI while revealing where demand actually travels. Owners could carry one unit between rooms; if they start taking it into cars or everywhere else, OpenAI gets a “real, useful signal” that a phone-like product might have a market. “It’s there when you want it. It’s not there when you don’t.”
Alexa’s success suggests the voice interface was stronger than the underlying assistant technology. Ben calls old-school Alexa “crappy” in an LLM world, yet its number-one use case still became music, solving real hands-busy problems—the hairdresser who can skip an overplayed song without stopping a haircut. For OpenAI’s device, “the most important component” may therefore be the microphones, not the speaker.
Hardware gives OpenAI a route around Apple, but it also narrows a service that ideally reaches everyone. Ben’s own objection is that hardware is “fundamentally constraining your market,” while Andrew notes that presenting a competitive threat could keep OpenAI at arm’s length. The counterargument: Apple is not going to let OpenAI replace Siri; if it did, that software position would be “obviously the best opportunity.”
The larger strategic contest is control of consumer touchpoints as models become more interchangeable. Ben argues that commoditization makes integration with the end-user experience essential, driving OpenAI and Anthropic toward software and into conflict with Microsoft and Apple. “There is a brewing fight for consumer touchpoints, and these companies are coming for it.”
OpenAI has not resolved whether it wants Apple-like hardware economics or Google-like service reach. Ben goes back and forth between pricing at cost and charging a premium, tentatively favoring premium pricing to test the market because early tech buyers will care less about price. His working assumption is “a few hundred dollars,” not Vision Pro’s “few thousand,” with giftability providing an unusually accessible adoption wedge.

Deep dive

1. OpenAI’s speaker makes hardware experimentation feel possible again

Andrew’s excitement is about resumed experimentation after what he sees as 5–10 years of limited hardware innovation, not certainty about the category. Once “deeply skeptical” of Vision Pro, he now uses it “five nights a week”; that reversal makes him eager for something new even if OpenAI’s hardware ambitions “may ultimately be crazy.”
Ben compares the AI moment to the 1990s, when newly online users realized they could build websites for small businesses. The appeal is that software feels open to tinkering again—“stuff that’s fun again.”
Gurman’s report describes a mobile, screenless companion for smart-home control, media, questions and messages. Ben’s “boombox” analogy isolates the crucial design choice: it runs on batteries and can follow its owner instead of remaining plugged into one room.

2. The home already proved that voice works despite weak assistants

Ben contrasts Amazon’s “single worst product” he has ever used, the Fire Phone, with the Echo launched three or four months later. The home gave voice an unusually coherent setting: phones may be charging, and people do not feel strange talking aloud when nobody else is around.
Alexa and similar home devices found a durable music use case despite assistants that now look remarkably limited beside LLMs. Their primary use ultimately became music, revealing both a durable voice habit and how little intelligence the original products supplied.
Ben’s best concrete example comes from a haircut: the hairdresser, hands occupied, tells Alexa to skip a song the staff has heard “57 times.” It is a small task, but one perfectly matched to hands-free interaction.
Recalling what he thinks was Steven Sinofsky’s observation, Ben says speakers mattered most in the old category; microphones will matter most now. A speaker listening to you sounds “weird and creepy” until the resulting product becomes useful enough to feel indispensable.

3. Portability doubles as an ambient-AI test

A battery-powered unit lets users carry AI between rooms rather than buy five stationary devices. Ben calls this a way to “fake it till you make it” toward ambient computing: “It’s there when you want it,” and not there when it is not wanted.
Behavior then becomes product research. If owners bring the device into cars or carry it everywhere, OpenAI learns whether demand points toward a phone; meanwhile, an add-on avoids asking consumers to replace their existing phone. Easy Wi-Fi setup also makes it a “great present,” and a failed V1 avoids going all the way into building a phone, cellular service and an ecosystem.

4. Apple makes OpenAI’s hardware strategy both necessary and dangerous

Ben applies his Meta critique to OpenAI: a service should reach everyone, while proprietary hardware “fundamentally” restricts its addressable market. The clean alternative would be letting other companies build devices while OpenAI supplies the AI service.
Andrew’s pushback sharpens the cost: challenging Apple could turn a potential distribution partner into a company that keeps OpenAI “at arm’s length.” Even before the current litigation, Ben wondered whether being the AI across Apple devices would be better than trying to replace the iPhone.
Ben’s answer rests on conviction that AI is a new paradigm Apple will not deliver and may obstruct. OpenAI remains behind Siri and must be deliberately invoked; replacing Siri would be “obviously the best opportunity,” but “that’s not going to happen.” Apple betting on itself therefore creates an opening for separate hardware.

5. Consumer ownership, not unit sales alone, determines the business

Ben’s broader thesis is that OpenAI and Anthropic have an economic imperative to own consumer touchpoints. As models become commoditized, each provider needs integration deep enough that customers cannot simply substitute another model—hence pressure on software, Microsoft and now Apple.
The unresolved choice is whether to price hardware at cost to expand the service or charge a premium and pursue hardware profits: “Are we trying to be Apple, are we trying to be Google?” Ben goes back and forth, but suggests starting premium to discover whether a market exists.
Andrew questions how a premium product wins initial share. Ben’s answer is a price-insensitive first cohort of tech enthusiasts, assuming the relatively simple microphones-and-speakers device costs “a few hundred dollars” rather than Vision Pro-scale thousands; broader economics can be played out over time after that market test.

Open Model Wars + Claire Stapleton's Dishy Google Memoir + Substack's Slop Fight Ep. 019 - Inside the STEEL Lab: From Package to Transistor (Teardown Lab)