Product Impact — News

Anthropic Is No Longer a Model Company

Arpy Dragffy — Thu, 09 Apr 2026 00:00:00 GMT

Anthropic Is No Longer a Model Company

Claude Managed Agents quietly redraws the competitive map for every AI infrastructure vendor.

By Arpy Dragffy · Founder, PH1 Research · Co-host, Product Impact Podcast April 9, 2026

Jessica Yan, a product lead at Anthropic, posted on LinkedIn yesterday to announce the public beta of Claude Managed Agents. Her post collected 261 likes, 21 comments, and 6 reposts in its first 21 hours. It is worth reading carefully because it quietly describes the most consequential strategic shift at Anthropic in 2026.

“You can now raise the ceiling of agent execution AND launch faster using our stateful APIs, performance-optimized harness, scalable infra, and rich developer tools.”

— Jessica Yan, Product at Anthropic

Read those four capabilities in order. Stateful APIs. Performance-optimized harness. Scalable infrastructure. Rich developer tools. This is not a model release. This is not a feature expansion. This is Anthropic announcing that it is in the agent platform business — and by extension, in direct competition with AWS Bedrock, Google Vertex AI Agent Builder, OpenAI Assistants API, LangChain, LangGraph, CrewAI, Dust, and every other piece of infrastructure currently hosting agent workloads.

Until Monday, building an agent on Claude meant you handled the infrastructure. Starting yesterday, Anthropic handles it for you.

That is a business model change, not a feature launch.

What actually shifted

Anthropic’s business on Monday was selling inference. You bought API access to Claude, you handled state management, you wrote the orchestration, you built the monitoring, you scaled the infrastructure, you owned the developer experience. The margin was inference margin. The customer was anyone running a workload.

Anthropic’s business on Tuesday is selling a platform. You get Claude and the infrastructure to run Claude-powered agents in production. Anthropic captures more of the value chain. The margin is platform margin. The customer is the developer building agent products.

Platform margins are higher than inference margins — that’s the obvious part. The less-obvious part is stickiness. An enterprise that builds its agent on Claude Managed Agents cannot easily port that agent to a competing model. State, tooling, operational patterns, and incident history all get locked into Anthropic’s infrastructure. Switching costs go up dramatically the moment a team’s agent is running on Anthropic’s harness.

This is the move the cloud providers have been waiting to see. It’s also the move they’ve been dreading.

Who gets structurally worse this week

The hyperscaler Claude distribution path. AWS Bedrock and Google Vertex host Claude for enterprises that don’t want to buy from Anthropic directly. Their value proposition is compliance, existing procurement relationships, and vendor consolidation. All three are real. None beat “the people who built Claude are also running your agent infrastructure for Claude.” Every enterprise that was going to run Claude agents through Bedrock or Vertex now has a reason to evaluate going straight to Anthropic instead.

The agent framework startups. LangChain, CrewAI, LangGraph, Dust, and a long list of others built their businesses on being the orchestration layer above multiple model providers. Their pitch was: don’t lock into one LLM; build with our framework; switch models when you need to. That pitch just got harder. Anthropic can now offer deeper integration, better performance tuning, and direct first-party support for Claude-based agents than any third-party framework can match. The frameworks will reposition around multi-model interoperability. That’s a harder sell than “we’re the best way to build agents.”

OpenAI’s Assistants API. OpenAI built Assistants to keep enterprise developers inside the OpenAI ecosystem. They will now have to respond to every Anthropic Managed Agents capability with an equivalent, while also fighting on ChatGPT Enterprise and the foundation model benchmark treadmill. OpenAI’s response will come fast. It will also be reactive, not strategic.

Who wins

Anthropic, obviously. They just expanded their addressable market from “developers buying model access” to “developers building agent products.” That’s a much larger number, at higher margins, with stickier customers.

The subtler winner is any enterprise that was paralyzed on build-versus-buy for its agent infrastructure. Managed Agents doesn’t eliminate the buy-side risk, but it gives risk-averse buyers a credible vendor-backed option they didn’t have on Monday. Expect the number of enterprises that move from “planning an agent platform strategy” to “piloting Anthropic Managed Agents” over the next 90 days to be larger than most analysts expect.

The question nobody in the coverage is asking

Here is what’s missing from every take this week: when a Claude Managed Agent takes a real-world action that causes a real-world problem, who is responsible?

The developer who wrote the agent? The enterprise that deployed it? Anthropic, whose platform is managing the state and executing the action? That question is not answered in Yan’s LinkedIn post. It is probably not answered in Anthropic’s initial documentation. It will be the first thing every enterprise general counsel asks before signing a contract, and it will be the single variable that determines whether Managed Agents gets enterprise traction or stays a developer tool.

Anthropic has two ways to handle this.

They can write a managed services agreement that places all liability on the customer. That’s legally clean and will scare off exactly the enterprises most likely to pay platform prices.

Or they can accept operational responsibility for the agents running on their platform. That solves the trust problem and fundamentally changes Anthropic’s risk profile as a company.

How Anthropic answers this question in their enterprise documentation over the next 30 days will tell you whether they see Managed Agents as a developer acquisition play or as a genuine enterprise platform. Those two paths lead to completely different outcomes in 2027.

Three things to watch in the next 30 days

Pricing. Anthropic has not published pricing for Managed Agents yet. Usage-based pricing signals developer targeting. Platform fee plus usage signals enterprise targeting. Whichever they pick will reveal who they’re actually selling to.

Named reference customers. The first three enterprise reference customers Anthropic cites will tell you whether they have enterprise credibility for this move. Watch the Anthropic blog through mid-May.

OpenAI’s response. OpenAI will ship something comparable within 60 days. They have to. How fast they respond — and whether it’s a feature match or a genuine platform strategy — will tell you how seriously they are taking this.

Yesterday Anthropic was a model company with a platform ambition. Today they are a platform company with a model at the center. The difference matters more than most of the coverage this week will capture.

Primary source: Jessica Yan, Anthropic — LinkedIn announcement (April 8, 2026)

About the author: Arpy Dragffy is founder of PH1 Research and co-host of the Product Impact Podcast. All claims about competitive positioning in this piece are based on public product documentation from the companies referenced.

The Man Who Hired Jony Ive Has a Warning for the Physical AI Boom

Arpy Dragffy — Thu, 09 Apr 2026 00:00:00 GMT

The Man Who Hired Jony Ive Has a Warning for the Physical AI Boom

Robert Brunner founded Apple’s industrial design group, built Beats, and shaped the original PowerBook. He thinks the AI hardware industry is making the same mistake all over again — and the next great tech companies will be the ones people trust with their lives, not just their data.

By Arpy Dragffy · Founder, PH1 Research · Co-host, Product Impact Podcast April 9, 2026

The 2026 race to put AI into a physical object is on, and the body count is already climbing.

Humane’s AI Pin — the most-hyped wearable launch of 2024 — was effectively dead by February 2025, when HP acquired its assets for $116 million after the company burned through more than $230 million in venture capital. Rabbit R1 is, by most accounts, on a similar trajectory. Meta quietly acquired Limitless in December 2025 and immediately stopped selling the Pendant to new customers. Meanwhile, OpenAI’s Chris Lehane confirmed at Davos in January that the company’s first device, designed in partnership with Jony Ive, is on track to ship in the second half of 2026 — a device OpenAI bought from Ive for $6.4 billion. Apple is preparing AI glasses, a camera pendant, and camera-embedded AirPods. Every major tech keynote at CES 2026 led with what the industry is now calling “physical AI.”

Almost nobody knows how to design for it.

The man best positioned to explain why showed up on the latest Product Impact Podcast episode, and what he said is more useful than anything that’s been written about the category this year.

Who is Robert Brunner

A short version of the résumé, because it matters: Robert Brunner founded Apple’s Industrial Design Group in 1989 and ran it until 1996. He hired Jony Ive (three times, by his own account, before Ive said yes). He led the design of the original PowerBook, whose keyboard-back, palm-rest pointing-device layout has remained the universal laptop configuration for 35 years. After Apple, he became a partner at Pentagram. In 2007 he founded Ammunition, the studio that designed Beats by Dre, the Square Stand, the Lyft Amp, the June Oven, the Polaroid Cube, and the Limitless Pin that Meta just bought.

Now, with co-founders, he is building Object — a startup focused on what physical AI should feel like when it’s designed to respect the user instead of extract from them.

When Brunner talks about how hardware should work, it is worth pausing the rest of the conversation and listening.

What he told us

Brunner’s central argument on the podcast is that the AI hardware industry is repeating the mistake the consumer software industry made fifteen years ago, but with a more dangerous payload.

“Modern technology is optimized for engagement, advertising, data extraction, time. In many ways, technology is, it’s like the matrix. It’s treating us as a source, as a resource. For information and not human well-being. And that’s one of the fundamental problems with digital technology. It’s been built around humans as a resource to be monetized. And I think we’re all sick of it.”

— Robert Brunner, Product Impact Podcast S02E06

The companies racing to put AI into wearables, pendants, glasses, and pins are, Brunner argues, building those products on top of the same incentive structures that made smartphones extractive. The hardware changes; the business model doesn’t. That’s the trap.

His framing of the alternative is the line worth tattooing on a product office wall:

“The most valuable currency in technology is rightfully becoming trust. The next great technology companies will be the ones people trust with their lives, not just their data.”

— Robert Brunner

He means this literally. As physical AI moves into devices that contain microphones, cameras, motion sensors, and access to an always-on data stream about how humans actually move through the world, the vendors that will win the next decade are not the ones with the best models. They are the ones whose customers genuinely believe the device is on their side.

Brunner’s test for whether AI in a product is real

This is the part of the conversation that should be compulsory reading for every product manager shipping an “AI-powered” anything in 2026.

Brunner offered a test for distinguishing genuine AI integration from AI-as-marketing-layer. It is short. It is brutal. It is the answer to a question every product team is being asked by their CEO right now.

“Does AI remove steps? Will the product require fewer actions to accomplish something meaningful — or more? If it adds menus and features and prompts and dashboards and all that stuff, it’s probably not good and it may just be marketing. But if AI quietly removes complexity and lets you do something faster, better, it’s real.”

— Robert Brunner

And then, the line that made me stop the recording:

“The best AI feature is the one you never notice. The problem simply disappears.”

This is the inverse of how every AI product release in 2026 has been marketed. Vendors are competing to show the AI — the chat overlay, the floating assistant, the “ask me anything” button, the badge in the corner of the interface. Brunner is saying that’s the tell. If you can see it, it isn’t working.

Compare this to what shipped with Humane’s AI Pin: a laser projector beaming a menu onto your palm, a wake-word interaction model, a visible badge on your chest that other people noticed before you did. The product made the AI as visible as possible. By Brunner’s standard, the design itself was the failure mode.

Why hardware is different from a chat interface

Brunner spent a long stretch of the conversation on something most coverage of AI hardware is missing: the relationship humans have with physical objects is fundamentally different from the relationship we have with software. He has been arguing this for thirty years. It is more relevant now than it has ever been.

“Human beings have this unique relationship with objects. In many ways we’ll use physical artifacts to define who we are — through the car we drive, the shoes we wear, the furniture we buy. People develop this emotional connection to things they can’t literally speak to, whether that’s a chair, a kitchen tool, whatever. That sort of goes back to the dawn of man — to when the first person who got up on two feet picked up a stick.”

— Robert Brunner

His point: when you put intelligence inside an object, you are not making the object smarter. You are inserting yourself into one of the deepest emotional relationships humans have with the made world. A chat interface is something you use. A wearable device is something you live with. The trust standard is dramatically higher, because the failure mode is so much more intimate.

This is the part of the analysis that the Humane and Rabbit failures keep teaching the market. Both products were technically functional. Both products had compelling demos. Both products lost their customers within months of shipping, and the postmortems keep finding the same root cause: users did not trust the device enough to live with it.

The Limitless example, in his own words

Brunner’s studio designed the Limitless Pin — the “memory augmentation” wearable that records audio throughout your day so an AI assistant can search and summarize it later. Meta acquired Limitless in December 2025. The pendant is no longer sold to new customers.

Brunner’s reflection on what went wrong is unusually direct for a designer talking about his own work:

“We chose to, instead of designing it to look like a piece of an iPhone or technology, we really designed it to be, feel more like a watch — a personal object — and came up with a really nice attachment system. But the fundamental challenge with the product, and essentially the product for those who don’t know about it, records audio. The fundamental issue is nobody wants to be recorded. Nobody. Even in meetings. And knowing that you’re being recorded — even though it’s got a little light that tells you that it’s on — you’re still like, okay, how is this information being used against me?”

— Robert Brunner

The form factor was right. The attachment was right. The model was right. The business was wrong because the product asked users to do something — let themselves be recorded all day — that no amount of design polish could make comfortable.

This is the diagnostic question Brunner is bringing to Object, the new startup he’s now building. It’s also the question every founder racing to ship a wearable in 2026 should answer before they tape out silicon.

The line OpenAI, Apple, and Meta should print and frame

Toward the end of the conversation, Brunner returned to the question of where AI will and won’t replace human contribution. His answer is the most succinct articulation I’ve heard of why the “AI replaces designers” thesis is structurally wrong:

“AI doesn’t feel. AI has never been hurt. AI has never felt joy. AI has never been through these experiences that shape you and define you. And those are the things that become these incredible assets — taste, insight, and judgment. Those are the things I think young designers need to spend more time developing, as opposed to learning how to do a specific tool or create amazing imagery. I don’t think those are things that will ever truly be replicated.”

And:

“Design can generate possibilities, but it can’t decide what matters.”

The product teams at OpenAI, Apple, Meta, and the dozens of physical AI startups currently raising rounds are about to discover this the hard way. The hardware will be impressive. The models will be impressive. The first generation will mostly fail anyway, because the people designing it will have optimized for what’s possible to demo instead of for what humans can actually live with.

The teams that survive the next 24 months will be the ones that take Brunner’s test seriously: does the AI make the product simpler, or does it make it noisier? Does the user notice it, or does the problem just disappear? Does the device respect the person, or does it extract from them?

The hardware boom is happening regardless of whether the industry takes that test seriously. Brunner’s bet, and the one his new company Object is being built around, is that the products that win 2027 will be the ones designed by people who already know the answer.

Listen to the full conversation: Product Impact Podcast S02E06 — Robert Brunner on Physical AI

Hosted by: Brittany Hobbs and Arpy Dragffy

About Robert Brunner: Founder of Ammunition, founder of Object. Former Director of Industrial Design at Apple (1989–1996). Hired Jony Ive. Designed the original PowerBook. Led design of Beats by Dre, the June Oven, Square Stand, Polaroid Cube, Lyft Amp, and Limitless Pin. (Wikipedia)

About the author: Arpy Dragffy is founder of PH1 Research and co-host of the Product Impact Podcast.

Sources used in this analysis: - Product Impact Podcast S02E06 (April 2026) — primary source for Brunner quotes - Axios: OpenAI aims to debut first device in 2026 - TechCrunch: Meta acquires Limitless - TechRadar: With the Humane AI Pin now dead - Robert Brunner — Wikipedia - Ammunition Group

The Year AI Leaves the Text Box

Arpy Dragffy — Thu, 09 Apr 2026 00:00:00 GMT

The Year AI Leaves the Text Box

Jensen Huang declared physical AI’s ChatGPT moment at CES 2026. Apple, OpenAI, Meta, and a wave of robotics startups are now racing to build the products. Almost nobody knows how to design for them — and the body count from the first wave is already growing.

By Arpy Dragffy · Founder, PH1 Research · Co-host, Product Impact Podcast April 9, 2026

The single most consequential announcement at CES 2026 wasn’t a product. It was a label.

When NVIDIA CEO Jensen Huang took the keynote stage in Las Vegas in January, he told the audience: “The ChatGPT moment for physical AI is here — when machines begin to understand, reason and act in the real world.”

Within 60 days, every major technology company had aligned around the same framing. Sam Altman confirmed in Davos that OpenAI’s first hardware device, designed in partnership with Jony Ive, would ship in the second half of 2026. Bloomberg reported on February 17 that Apple is “ramping up work on glasses, a pendant, and camera AirPods for the AI era,” confirming three simultaneous wearable categories that Tim Cook has personally championed. Mark Zuckerberg told TechCrunch on January 28 that “a future without smart glasses is hard to imagine,” and Meta is now in talks with EssilorLuxottica to double Ray-Ban smart glasses production to 20 million units by the end of 2026 — potentially climbing to 30 million if demand holds.

Three years after the AI revolution began inside a text box, 2026 is the year it leaves the text box. The most-funded category in tech is now the race to put AI inside an object you can wear, carry, or live with.

There is just one problem. The first wave of physical AI products has already mostly failed — and the people who actually know how to build successful consumer hardware are quietly warning that the second wave is on track to fail in the same ways.

What the leaders are saying

The unusual feature of this moment is that the heads of every company racing into physical AI are publicly aligned on what’s happening. The disagreement is on what to do about it.

Jensen Huang, NVIDIA CEO, framed it at CES 2026 as a category-defining inflection. “Breakthroughs in physical AI — models that understand the real world, reason, and plan actions — are unlocking entirely new applications,” he said. NVIDIA’s bet is the picks-and-shovels position: it doesn’t intend to build the robots or the wearables, it intends to own the compute and the foundation models that everyone else builds on top of. NVIDIA’s CES announcements included the Cosmos physical AI simulation framework, the Alpamayo automotive reasoning model, and the Rubin chip platform shipping to partners in the second half of 2026.

Sam Altman, OpenAI CEO, has been more guarded. According to leaked internal conversations reported by TechCrunch, Altman told staff that OpenAI plans to ship 100 million devices “faster than any company has ever shipped 100 million of something new before.” Publicly, he has been more measured. In November he described the device as something that should feel “more peaceful and calm” than a smartphone. About an earlier prototype, he told reporters: “There was an earlier prototype that we were quite excited about, but I did not have any feeling of: ‘I want to pick up that thing and take a bite out of it.’”

The OpenAI device, originally branded “io” through the $6.4 billion acquisition of Jony Ive’s design firm, has now been delayed beyond 2026 due to privacy, compute, and “personality” issues, according to Windows Central reporting on Altman’s most recent comments. “Do not expect anything very soon,” Altman reportedly said. The brand also had to be changed because of a trademark dispute with hearing-aid startup iYo.

Mark Zuckerberg, Meta CEO, has been the most aggressive in his framing. He told TechCrunch on January 28 that he believes smart glasses will be ubiquitous within the decade. “It’s hard to imagine a world in several years where most glasses that people wear aren’t AI glasses,” he said. Meta’s Ray-Ban Display launched in late 2025 at $799, bundled with a Meta Neural Band — an EMG wristband that translates muscle signals into commands. The glasses are being rolled out across France, Italy, Canada, and the UK in early 2026. Meta has reported that smart glasses sales tripled in 2025, with more than seven million pairs sold.

Tim Cook, Apple CEO, has been the quietest in public — but Bloomberg’s reporting describes the AI pendant, the camera-equipped AirPods, and Apple Glass as “Tim Cook’s top priority products.” 9to5Mac reported in late 2025 that Cook is personally driving the program as Apple’s third major product category after the iPhone and the Apple Watch. Apple Glass is targeting late 2026 unveiling with shipping in 2027, according to Bloomberg’s sources, and will not initially include AR functionality — instead positioning itself as an iPhone accessory anchored on Apple’s Visual Intelligence system.

Four CEOs. Four different bets. One agreement: the next decade of AI value lives outside the chat interface.

What’s actually shipping (and when)

The product roadmap for the next 18 months, drawn from publicly-announced timelines and verified reporting:

Already shipping or imminent: - Meta Ray-Ban Display — $799 AI glasses with full-color display and EMG wristband, available now in the US, expanding to Europe and Canada in early 2026 - NVIDIA Rubin platform — physical AI compute infrastructure, shipping to NVIDIA partners in the second half of 2026 - Figure 03 humanoid robot — designed for high-volume manufacturing, introduced late 2025 (Figure AI has raised over $1 billion to date) - Apptronik Apollo humanoid robot — Apptronik raised $520 million in February 2026 at a $5 billion valuation, with explicit ambitions to beat Tesla Optimus to market

Expected H2 2026 (per public statements): - OpenAI’s first device — originally “io,” now under a new name pending trademark resolution; a screen-free, possibly behind-the-ear wearable codenamed “Sweetpea”; recently delayed and may not actually ship until 2027 - Apple Glass — late 2026 unveiling, 2027 shipping per Bloomberg sources - Apple AI Pendant — pinned-to-shirt or worn-as-necklace form factor, late 2026 development push per Bloomberg

Expected 2027 or later: - Camera-equipped AirPods — late 2026 to 2027 per Bloomberg - Tesla Optimus Gen 3 — Elon Musk has set 2026 as the year Optimus moves to higher-volume external sales, with a target price of $20,000 to $30,000 per unit; production has been delayed multiple times

Already dead or absorbed: - Humane AI Pin — HP acquired the assets for $116 million in February 2025 after Humane burned through more than $230 million in venture capital - Limitless Pin — Meta acquired Limitless in December 2025 and immediately stopped selling the Pendant to new customers - Rabbit R1 — by early 2026, reports of unpaid employee salaries and a 1.5/5 Android Authority rating suggest the company’s runway is running out

That’s the most-funded, most-hyped, most-public hardware race in technology since the smartphone era. The first wave is mostly already in the graveyard. The second wave is being designed right now.

The warning from the man who hired Jony Ive

The most useful analysis of why the first wave failed — and the most credible warning about the second — came not from a tech executive but from the designer who built the playbook the entire industry is now trying to apply.

Robert Brunner founded Apple’s Industrial Design Group in 1989. He hired Jony Ive (three times, before Ive said yes). He led the design of the original PowerBook, whose layout has remained the universal laptop configuration for 35 years. After Apple, he founded Ammunition, the studio that designed Beats by Dre, the Square Stand, the June Oven, the Polaroid Cube, the Lyft Amp — and the Limitless Pin that Meta just acquired. He is now building a startup called Object focused specifically on what physical AI should feel like when it’s designed to respect users instead of extract from them.

Brunner joined the Product Impact Podcast in early April, and his diagnosis of the category should be required reading for every founder racing to ship before OpenAI does.

“Modern technology is optimized for engagement, advertising, data extraction, time. In many ways, technology is, it’s like the matrix. It’s treating us as a source, as a resource. For information and not human well-being. And that’s one of the fundamental problems with digital technology. It’s been built around humans as a resource to be monetized. And I think we’re all sick of it.”

— Robert Brunner, Product Impact Podcast S02E06

Brunner’s argument: the AI hardware race is repeating the mistake of the consumer software industry, but with a more dangerous payload. The vendors are betting that putting intelligence inside a wearable will produce a new category of product. Brunner is betting that without a fundamentally different relationship with the user, the form factor doesn’t matter.

His test for whether AI in a product is genuine or marketing:

“Does AI remove steps? Will the product require fewer actions to accomplish something meaningful — or more? If it adds menus and features and prompts and dashboards and all that stuff, it’s probably not good and it may just be marketing. But if AI quietly removes complexity and lets you do something faster, better, it’s real.”

“The best AI feature is the one you never notice. The problem simply disappears.”

Compare this to what shipped with Humane’s AI Pin: a laser projector beaming a menu onto your palm, a wake-word interaction model, a visible badge on your chest. The product made the AI as visible as possible. By Brunner’s standard, the design itself was the failure.

And on the trust question that nobody in the industry is solving:

“The most valuable currency in technology is rightfully becoming trust. The next great technology companies will be the ones people trust with their lives, not just their data.”

Brunner is unusual because he is willing to talk about his own studio’s failures. On the Limitless Pin specifically — the product Meta just bought and pulled from sale — he was direct: the form factor was right, the attachment system was right, the AI worked. The fundamental issue, in his words, was that “nobody wants to be recorded.”

The implication for every wearable currently in development at OpenAI, Apple, Meta, and the dozens of startups racing to ship: the design is not the moat, the model is not the moat, the form factor is not the moat. The moat is whether your customer is willing to put your device on their body in 2027.

What to watch in the next 90 days

Three things will determine whether the second wave of physical AI is a category breakthrough or a more expensive repeat of the first.

OpenAI’s launch timeline. Altman’s reported “do not expect anything very soon” walks back the H2 2026 ship date Lehane gave at Davos. Whether OpenAI ships in 2026 or slips to 2027 will signal whether the company has solved the fundamental design problems Brunner identified — privacy, “personality,” trust — or simply pushed them down the road.

Apple’s reveal. The Apple Glass unveiling, expected late in 2026, will be the moment Apple’s bet on Visual Intelligence becomes real. Apple has the credibility and the supply chain to ship at scale. Whether the first product ships with cameras-on by default, and how Apple frames the privacy posture, will set the standard the rest of the category has to match.

Meta’s production volumes. If Meta hits the 20 million Ray-Ban units it’s targeting for 2026, the smart-glasses category will have already won the volume war before OpenAI or Apple ship anything. If Meta misses, the entire premise that wearables are a mass-market AI category gets called into question.

The thing every CEO has stopped saying out loud, but every product team should be discussing: the first wave failed not because of bad models, bad chips, or bad form factors. It failed because users decided, individually, day by day, that they did not trust the device enough to live with it.

Brunner’s line is the one to leave with:

“AI doesn’t feel. AI has never been hurt. AI has never felt joy. AI has never been through these experiences that shape you and define you. And those are the things that become these incredible assets — taste, insight, and judgment.”

The companies that build the trillion-dollar physical AI market of 2030 will be the ones that figure out how to put taste, insight, and judgment into the design — not just the model.

The body count of the first wave suggests that may take longer than the keynote slides imply.

Listen to the full Brunner interview: Product Impact Podcast S02E06 — Robert Brunner on Physical AI

About the author: Arpy Dragffy is founder of PH1 Research and co-host of the Product Impact Podcast.

Sources used in this analysis (all linked inline above):

Jensen Huang CES 2026 keynote — Axios, January 5, 2026
Sam Altman device timeline — Axios, January 19, 2026; TechCrunch, January 21, 2026
OpenAI device delay — Windows Central, recent
Apple AI hardware roadmap — Bloomberg, February 17, 2026
Tim Cook’s “top priority” framing — 9to5Mac, December 22, 2025
Mark Zuckerberg on smart glasses — TechCrunch, January 28, 2026
Meta Ray-Ban Display launch — Meta blog, September 2025
Meta production volume targets — Fintool News, recent
Humane AI Pin acquisition — TechRadar, February 2025
Meta acquires Limitless — TechCrunch, December 5, 2025
Apptronik funding — CNBC, February 11, 2026
Robert Brunner background — Wikipedia; Ammunition Group
Robert Brunner quotes — Product Impact Podcast S02E06, April 2026 (primary source)