BRETT ADCOCK has a habit of starting companies that sound implausible and then raising eye-watering sums to will them into existence. He sold his first startup, a recruiting marketplace called Vettery, to Adecco for $100 million. He co-founded Archer Aviation and took it public at $2.7 billion. Then came Figure AI, a humanoid robotics firm that raised over $1 billion last year and now carries a $39 billion valuation — making Adcock, a farm kid from central Illinois, worth an estimated $19 billion according to Forbes. Now he has started Hark, a new AI lab seeded with $100 million of his own capital, and the pitch is characteristically grand: build the most advanced personal intelligence on the planet, paired with bespoke hardware designed to serve as the interface between humans and machines.
Hark emerged from stealth last week after eight months of quiet hiring. The lab has poached over 30 engineers from Google, Meta, Amazon, and — most notably — Apple, where it recruited Abidur Chowdhury, the industrial designer behind the iPhone Air, as head of design. Adcock plans to scale the team to 100 by mid-year and ship Hark's first AI models by summer. The company is building end-to-end speech, text, and vision models with what it calls "highly personalized memory," alongside a series of native hardware devices. Adcock will remain CEO of both Hark and Figure simultaneously, a dual role that has raised eyebrows even among admirers. The ambition, per an internal memo reported by The Information, is to create AI that can "think proactively, recursively improve, and care deeply about people."
Ghosts in the machine
But the last company to promise a post-smartphone AI device paired with a proprietary model was Humane — and that story ended with a $116 million fire sale to HP and bricked devices on users' chests. The Rabbit R1, another AI hardware darling from the same era, shipped 100,000 units in pre-orders only to see active users collapse to roughly 5,000 within months. Both ventures failed for the same structural reason: they asked consumers to carry a second device that did less than the phone already in their pocket. The graveyard of dedicated AI hardware is fresh enough that even sympathetic observers might wonder what, exactly, Adcock thinks he knows that Humane's former Apple executives did not.
The answer, Adcock reckons, is embodiment. His critique of the current AI paradigm is pointed and specific: large language models, he has argued, are essentially "advanced Google search engines" that lack the world-understanding necessary to avoid walking through glass walls or crushing delicate objects. This is not merely a rhetorical flourish — it is the thesis that led Figure to dissolve its partnership with OpenAI in 2025, with Adcock claiming his team "ran circles around them" on robotics-relevant AI. Hark, in this framing, is not a consumer gadget play. It is the cognitive layer for a future in which AI systems see, hear, touch, and move through physical space. The hardware is not a phone replacement; it is, in theory, a bridge to the embodied intelligence that Figure's humanoids will eventually require.
The strategic logic is formidable on paper. If the most valuable AI stack of the coming decade is not a cloud-hosted chatbot but a continuously learning, physically grounded intelligence — and if the entity that controls both the hardware deployment network and the model training pipeline will enjoy extraordinary structural advantage — then building Hark alongside Figure is not distraction. It is vertical integration. Adcock is, in effect, betting that the model layer and the physical layer are inseparable, and that training them jointly will yield capabilities neither could achieve alone. His brother Colby, for what it is worth, runs Scout AI, a defense startup building autonomous military systems; the Adcock family's collective wager on embodied intelligence is nothing if not comprehensive.
Still, the execution risk is considerable. Running two frontier companies is the kind of arrangement that works brilliantly until it doesn't — and the history of dual-CEO structures in technology (ask anyone who watched Jack Dorsey toggle between Twitter and Square) is not uniformly encouraging. Hark's first models are months away. Its hardware roadmap is undisclosed. And the AI industry's capacity to absorb $100 million without producing a shipping product is, at this point, well documented. Adcock has earned the benefit of the doubt through sheer serial improbability; the question is whether personal intelligence is the kind of problem that rewards a founder's conviction, or the kind that has already swallowed several hundred million dollars in other people's confidence. The summer release will begin to tell. The humanoids, presumably, will be watching. ■
For more, join 75,000 subscribers getting tech's favorite brief here
