The most undervalued moat against AI in consumer hardware: past failure.
Not startup failure. Product failure. The kind you can only earn by trying to make something that has never been made before.
Hearing fourteen factories say it can’t be done before one says, “I think I have a way to do this.” Watching $85K in tooling get scrapped because a geometry that was perfect in simulation couldn’t survive production. Holding a customer return where the material you spent months testing cracked after ninety days in someone’s pocket.
AI can generate a thousand product concepts in an afternoon. It will never generate this. And the people who carry this knowledge, in their hands, in their notebooks, in their scars, are about to become the most valuable talent in hardware.
The knowledge that doesn’t exist in any dataset
The Dyson engineer who worked on a hundred inventions that never shipped knows exactly why your motor housing will fail thermal testing. That’s not in any patent. It’s not in any training set. The Nike materials scientist who knows a specific foam compound will lose 30% of its energy return after 200 miles? That knowledge exists in maybe a dozen heads worldwide.
This is the part of product design AI is structurally bad at, and will remain bad at for longer than the headline AI-design demos suggest. Training data flattens distinctions that matter at the bench. A model that has read every patent and every product review knows that high heels and hiking boots are both footwear. It does not know that the senior designer who has shipped 11 SKUs of mountaineering boots over 14 years has a feel for the sole compound durometer that no amount of fine-tuning will reproduce.
And the specificity goes deeper than most people realize. “Footwear designer” is not a useful category. The person who designs a rubber mountaineering boot for REI’s customer has almost nothing in common with the person who designs a 110mm stiletto for Louboutin. Different materials science. Different last construction. Different customer psychology. Different price elasticity. Different failure modes. Different factory relationships in different countries. The one is solving for waterproof seams at −20°C; the other is solving for whether the heel will catch a Manhattan sidewalk grate. Calling them both “footwear designers” obscures more than it explains.
The same fractal applies inside hardware. The senior designer who shipped a stick vacuum for apartment-dwellers at Dyson is not the right person to design a contractor-grade shop vac, even though the org chart at most companies would file them under the same job title.
AI will get better at simulating some of this. Generative design and AI-CAD are already eating the commodity work — rendering, parametric iteration, geometric variation. The mid-tier industrial designer whose value was “I can produce competent geometry” is in serious trouble, and the industry hasn’t fully reckoned with it yet. But the judgment layer — the category-specific, failure-earned, physically embodied knowledge that determines whether a product succeeds or fails — is a fundamentally different kind of skill. And it’s the one that matters.
What happens when that knowledge meets the factory floor
Non-public knowledge is only half of the story. The other half is what the person who carries it actually does.
I’ve been there: a product that’s never been made before, a dozen factories declining to even try, months finding the one partner willing to develop custom tooling from scratch. Failing. Rebuilding. Failing again. Finally creating something nobody thought was possible, because the designer in the room had the judgment to know which failure to push through and which to walk away from.
That’s the real job. Not renders. Not CAD. Taking a napkin sketch through a hundred collisions with reality: engineers, factories, customers, physics, and holding the thread until it becomes something people wear every day, put in their kitchen, build their lives around. Absorbing the failures, walking into the engineering room to explain why the form factor has to change, walking into the customer meeting to hear what people actually need, then going back to the bench and reconciling all of it into something that works.
That full loop: invent, test, fail, listen, bridge, iterate, ship — is what turns hardware startups into companies that matter. No AI on earth can run that loop. Not this decade.
The founding-team math is changing
Through the 2010s, the canonical consumer hardware founding team was technical co-founder plus business co-founder. Engineering was the scarce skill; design was a hire you made later, after Series A, when you could afford it. That assumption is breaking.
As AI commoditizes the engineering execution layer, the scarce skill is shifting to design judgment: the failure-earned, category-specific kind. And design judgment is structurally harder to hire post-facto than engineering ever was, because the senior people with real category-specific expertise don’t take design-lead jobs at unproven startups. They want equity, founding status, or nothing.
The companies getting this right are increasingly being founded with a design co-founder from day one.
Oura is the obvious example. Kari Kivelä, the jewelry-trained co-founder, was the reason the ring looked like jewelry instead of a medical device. Strip him out of the founding team and Oura is a failed startup with great sensor tech. (More on this in the next post.)
Carl Pei’s Nothing made the design-led founding thesis explicit, and the early hardware reflects it: the products are aesthetic statements first and feature-comparable second, which is exactly backwards from how engineer-founded companies usually sequence those decisions.
While not from the Nordics, Dyson is the same story. James Dyson was the designer-founder. The engineering followed the design conviction, not the other way around.
Teenage Engineering was founded by industrial designers, not engineers. The whole company is an argument for design-first founding teams, and the pricing power on a $1,499 portable synthesizer reflects it.
For every one of those, there are a half-dozen cautionary tales, Pebble, Jawbone, Juicero, and the long graveyard of post-Kickstarter wearables, where the engineering was credible and the design judgment wasn’t, and the products got progressively less wantable until they didn’t sell. In each case, the founding team had no design partner with real authority, and design got hired in to execute decisions that should have been made by a co-founder.
The matchmaking problem
The good news: many of those senior designers and engineers have left the Dysons and Nikes. They’re out there. Some are consulting. Some are between roles. Some are open to the right founding opportunity.
The hard part is finding the one whose scars match your specific product. That matchmaking — the right expert for the right category at the right moment — is becoming one of the most important problems in consumer hardware.
For founders without a design co-founder, the standard options aren’t great. Hiring senior designers from Apple, Dyson, or Whirlpool can be really difficult for a pre-Series-A startup — those people typically have stock vesting, families, and no appetite for early-stage risk. Agencies tend to be generalists who claim category expertise they don’t have. The category-specific senior expertise you actually need is locked inside a handful of incumbents, mostly under NDA.
A few firms are emerging that try to bridge this gap by matching founders to specific senior designers and engineers from tier-one companies in their exact product category. Gembah, where I’m involved as a founder and board member, is one model; there are others worth watching. The market is early, and the need is growing faster than anyone expected — precisely because AI is making every other part of the product development process easier, which makes the part it can’t do (failure-earned judgment) more valuable by contrast.
The honest counter
AI eventually gets better at simulating category-specific design judgment too. A model trained on every Dyson internal QA report, every failed prototype, every supplier email about a tooling failure could approximate the senior designer’s feel. That day is coming. It’s not here yet, and the gap between now and then is, coincidentally or not, about seven years (more on that number in a later post).
Until then, the founders who treat design judgment as a founding-team problem rather than a hiring problem will compound advantages the rest will spend years trying to backfill. The next decade of consumer hardware winners will look more like Oura and Nothing than like the engineering-led startups that defined the 2010s. And the investors who internalize this earliest will see it in the cap tables before they see it in the products.
This is the first in a series of posts on what’s actually changing in consumer hardware as AI reshapes the stack. Next: how Oura’s jewelry-trained co-founder is the reason that company is worth $11 billion and not a footnote.