The AI Assistant Era

Musings on OpenAI hardware rumors

“Hey Siri, use ChatGPT…”

Siri, a startup Apple bought for roughly $200M, first launched on the iPhone in 2011.

Today, that price might look like a steal for a major incumbent buying an AI company. But the Siri experience, under Apple’s roof, has remained fundamentally unchanged for nearly 15 years. Now the iPhone’s architect Jony Ive works for OpenAI. And if you ask Siri a complex question, it’ll ask if you’d like to use ChatGPT to get the answer.

I’ve been surprised there hasn’t been more news on what’s coming this year in hardware land. OpenAI announced it acquired Jony Ive’s “io Products” for ~$6.5B in stock six months ago. Since then, ChatGPT added another ~100M users by most estimates, bringing its total user base to just under a billion. For context, Apple’s total installed base of iPhones is roughly 1.5B, which represents about one in every five people on the planet. This compares to one in every nine or ten humans using ChatGPT.

Comparing the number of ChatGPT users with Apple’s iPhone installed base is somewhat of a fool’s errand. Apples and oranges, one might say. Especially since only ~5% of OpenAI’s users are paying customers and ChatGPT is software and not hardware.

Still, a consumer application achieving 1B users in three years is wild. How wild, you ask? Instagram took eight years to reach 1B. TikTok took five years.

From Innovation to Artifact

Devices like Google Home and Apple HomePod already felt dated to me, even before ChatGPT. Now they feel like relics from another era.

For example, my favorite restaurant in Miami is Sunny’s. I go often, and each time I use Apple CarPlay to navigate there. On the first attempt, Siri always finds some other place called “Sonny’s” which is six hours away — impractical to say the least. (And separately, Sonny’s looks like somewhere I’d never go.) I have to spell out “Sunny’s, S-U-N…” in order for maps to find the right location. I use Google Home to turn my lights on and off and check the weather. That’s not enough to justify a replacement cycle.

At the same time, I can open ChatGPT and ask the AI coach I created — simply by prompting and using memory — in-depth questions about my life, relationships, and existence on this planet. The answers it gives back are not only good, they’re insightful.

Even just sticking with the Maps use case, ChatGPT was also able to remember the name of a cool nightclub I went to in my twenties that no longer exists. My friends in NYC were reminiscing on old times over the holidays and I couldn’t remember Westway. (RIP.)

Exhibit 1: A real Q&A between me and ChatGPT about Westway

Me: “What was the name of the club on the west side highway that was setup like a strip club but wasn’t. In NYC.”

ChatGPT: “You’re likely thinking of The Box or Scores, but given your specific clue—“set up like a strip club but wasn’t”—the most fitting answer is probably The Westway. Here’s the breakdown.

• The Westway (2011–2015):

  • Located on the West Side Highway, in the Meatpacking District.

  • It was literally built in a former strip club, and kept some of that aesthetic—mirrors, poles, velvet booths—but was primarily a nightclub and performance venue, not an actual strip club.

  • Hosted wild parties, music events, and fashion week after-parties.

  • Attracted models, downtown creatives, and celebrity nightlife crowds.

Let me know if you’re thinking of a different era or vibe—there are a few other niche spots like Bungalow 8 or Veruka, but Westway sounds like the closest match based on your description.

But Siri can’t find Sunny’s? There’s a tragically massive disconnect between these two experiences and I believe it’s about to converge very soon.

Ive, who worked at Apple for 27 years — most recently as Chief Design Officer and widely considered the architect behind the iPhone — is leading the industrial design charge on what’s described as a family of AI devices at OpenAI: simple, unobtrusive objects meant to fade into the background rather than demand attention.

Based on a Perplexity search I ran this morning compiling the latest news, rumors point to OpenAI shipping their first device in late 2026 or early 2027. According to multiple leaks, they’ve completed initial prototypes of a pocket-sized, screenless device. Something like an iPod Shuffle. It’s audio-first, built around microphones, speakers, and possibly cameras, designed for ambient, context-aware assistance.

It also appears, based on news, that OpenAI is reorganizing internally around audio, consolidating engineering and product groups to improve next-gen voice models with more natural speech, interruption handling, and overlapping dialog—explicitly linked to the upcoming hardware. The hardware is expected to run OpenAI’s latest frontier models, ingest continuous environmental data (audio, visual) via onboard sensors, and sync with phones and laptops rather than replace them, acting as a “third core device.”

My Predictions

I‘ve met with over twenty founders building AI assistants. So far I’ve invested in one (Jack) and spent six months building alongside the founding team on another in the coaching and expert advice space. As I write this, OpenAI announced it acquired Torch for $100M, marking a deeper push into healthcare. None of this is about hardware specifically, but it shapes my view on what the device will and won’t do initially.

I like the idea of a device that helps us be more mindful and off-screen. But will we trade screens for talking to a robot clipped to our shirt? Based on gut instinct above all else, here are my predictions on OpenAI’s hardware:

1. Use cases will follow ChatGPT’s adoption curve

I think OpenAI’s hardware will start with the core use cases people already rely on ChatGPT for: answering questions, explaining concepts, and capturing ideas through voice. The device will be ChatGPT in your pocket — listening, ready to help you think through problems or take notes while you’re walking, cooking, or commuting. From there, it will layer in services. Email and text composition. Calendar management. Food delivery through Uber Eats. Smart home control. Navigation and local recommendations… it will know where you are and what’s nearby. It won’t try to do everything on day one, but the path is clear with open APIs, etc.

2. Startups and vertical-specific use cases

I think OpenAI will face competition not from Google or Amazon, but from startups building for niche use cases: home management, childcare, education, elder care, etc. I also think that in many cases, these vertical players will win. The large infrastructure players can’t optimize for every context, and they shouldn’t try. The wedge is specificity.

3. Health data integration, not competition

I don’t think OpenAI will compete directly with Oura or Whoop, but I do think the OpenAI device will retrieve and output your health data on-demand. Voice query, instant answer. No app to open, no screen to unlock. The device becomes the interface layer for all your wearables — ask it about your sleep score, heart rate variability, or whether you’re ready for a hard workout. The data stays with the specialized hardware companies who do measurement well, but OpenAI owns the conversational access layer.

4. Voice-accessed app store

I think the device will access OpenAI’s app store entirely by voice. “Install the Montessori learning assistant.” “Open my meal planning app.” The apps won’t be visual; they’ll be conversational. This is the natural evolution of what we’re already seeing with ChatGPT plugins and GPTs. Developers will build specialized agents that run on OpenAI’s hardware, and users will discover them through natural language rather than scrolling through categories. The business model could write itself: OpenAI takes a cut of subscriptions, which startups will pay for distribution, just like Apple did with native iOS apps.

5. Three colors, just like the iPhone

And finally, I think the device will come in three colors, just like the iPhone. Ive knows that hardware isn’t just about function — it’s about identity. People will want choice, even in something meant to disappear. This seems trivial until you remember that color options signal that this is a consumer product, not a tech prototype. It’s a statement about mass-market ambition. (See OpenAI’s most recent ad campaign directed at consumers.)

Exhibit 2: Mike’s ChatGPT generated mock-up of OpenAI’s hardware

Source: Mike prompting ChatGPT

The Open Question

If not skeptical, I’m curious about adoption. Are people actually going to walk around talking to devices all day? Voice has always promised more than it delivered IMO. We’ve long had phone calls. We have voice notes. Yet text remains the dominant interface.

Everyone is about to have an AI assistant. What that assistant will actually do for us, and whether we’re ready to talk to it, is up for debate. The real question is whether OpenAI can create the first major platform shift since the iPhone launched in 2008.