Anthony Franco

Why Design Just Became AI's Most Expensive Ingredient

In 2025, OpenAI acquired io, a hardware company Jony Ive co-founded, in a deal valuing it at about $6.5 billion.

Not a model company. Not a chip company. Not a data company. A hardware company. Jony Ive’s io Products, the hardware company he co-founded after leading design work on the iPhone. Ive took on design and creative responsibilities across OpenAI while remaining independent. The prototype was called “the coolest piece of technology the world will have ever seen.”

This is the most important signal in AI right now. And most people missed it because they were arguing about benchmarks.

The Graveyard Next Door

The same year OpenAI acquired io, the AI hardware graveyard kept filling up.

The Humane AI Pin (launched at $699) was a disaster. The Rabbit R1 ($199) was a paperweight. Both had incredible technology. Both had massive hype. And both failed for the same reason: they were technology in search of a problem.

OpenAI made its largest acquisition to date. Not for better AI. For better design.

That should tell you something.

The Interface Gap

I spent a decade building one of the first premier UX firms in the country (EffectiveUI). We worked with 40% of the Fortune 100 during the mobile revolution. I saw the same pattern then that I see now.

Engineers build what is possible. Designers build what is necessary.

Right now, AI is in its “Command Line” era. We’re typing text into a blank box and hoping for the best. It’s powerful, yes. But it’s high-friction. It demands that the user be an expert in the machine’s language (prompt engineering) rather than the machine understanding the user’s intent.

To me, Jony Ive’s lesson from the iPhone wasn’t a better processor. It was an interface my grandmother could use.

He made the technology invisible.

Design is Not Decoration

Most corporate leaders think design is “making it pretty.” They think it’s the coat of paint you apply at the end.

This is wrong. Design is the architecture of the interaction.

In the context of AI, design is the answer to three questions:

  1. Prediction: How does the system know what I want before I ask?
  2. Correction: How do I tell it when it’s wrong without starting over?
  3. Trust: How does it prove it did the work?

The Rabbit R1 failed because it ignored the Correction loop. WIRED reported inaccurate answers. The Humane Pin failed because it ignored the Trust loop. It spoke answers aloud, and users could review their query history in Humane Center.

Look at what we know about the OpenAI device. Reportedly screenless, with audio and visual sensing. Reportedly, no screen. If that product gets even one of those three questions wrong, it’s another Humane Pin.

Reports suggested development problems could delay its release. The Financial Times reported privacy, computing-power, and assistant-personality problems. Translated into design language: Trust isn’t solved. Prediction isn’t fast enough. And the Correction loop for a screenless device is a problem nobody has cracked yet. I see design problems here as well as technical ones.

Why Observation Comes Before Building

These failures, the Pins, the Rabbits, and likely the first wave of whatever comes next, share a root cause: the teams built from technical capability instead of user experience. The question should be “What does a person actually need in this moment?” instead of “What can our AI do?”

This is what AI First Principle #6 gets at: Design systems from lived experience, not distant observation. The people wrestling with system failures are the ones qualified to design system futures. Not the engineers in the lab. Not the executives in the boardroom. The people doing the work.

The Humane team spent years developing a wearable with a projector. In my view, audio-only answers make verification harder.

Observation first. Technology second. Every time.

The $6.5 Billion Lesson for Your Business

You probably aren’t building consumer hardware. But you are building AI workflows for your team or your customers.

And you are probably making the same mistake Rabbit made.

Suppose you’re focusing 90% of your budget on the “Model.” Which LLM to use, how to fine-tune it, where to host it. You’re focusing 10% on the “Interface.” How the human actually interacts with it.

Flip that ratio.

The model is a commodity. GPT-4, Claude, Gemini, they’re all brilliant commodities. The difference between a successful implementation and a failed one is design.

OpenAI chose to acquire a hardware company. They bought io and gave Ive design and creative responsibilities across OpenAI. What does that investment in hardware and design tell you about your own implementation?

The “Invisible” Standard

The ultimate goal of AI design is to make the AI disappear.

When you use Spotify, you don’t say, “Wow, what a great recommendation algorithm.” You just say, “I love this song.” When you use Uber, you don’t say, “The routing optimization is impressive.” You just say, “My car is here.”

That’s the goal I see in this project. Not a “better AI device.” A device where you forget you’re using AI at all. I think of it as calm computing.

If your users have to “think about the AI,” you haven’t finished designing it. And if you think that’s easy, consider the development problems reported by the Financial Times.