Episode 851 of My First Million. Sam Parr and Shaan Puri interview Brett Adcock, founder of Figure (humanoid robots), Hark (a new AI lab building an AI that uses computers like a human), Cover (weapon detection for schools), and previously Archer Aviation and Vettery. Adcock walks through why he believes AI will be 100 times bigger than the internet, why the phone and the laptop are the wrong interface for AI, and why intelligence, not manufacturing, is the binding constraint on humanoid robots. The hosts then audit the predictions he published on January 1st, 2026, one by one, and he grades each on track or slipping. He also details the AI talent market from the inside, including a $36 million Meta counteroffer that beat his own, and the fifteen years of near bankruptcy that preceded any of it.
Published Aug 14, 202658:10 video64 min readAdded Aug 27, 2026Open on YouTube →
At a glance
The thumbnail says "a $39B tech founder" and never names him. It is Brett Adcock, founder of Figure, the humanoid robot company, on his third annual visit to My First Million with Sam Parr and Shaan Puri. He also runs Hark, an AI lab about a year old that is building an AI which drives a computer the way a person does, and Cover, a weapon detection company aimed at school shootings, and before those he took Archer Aviation public and sold the recruiting marketplace Vettery for $110 million.
The hook is an audit. On January 1st, 2026, Adcock published a list of predictions for the year. Eight months in, Sam reads them back and makes him grade himself, live, on track or off. He goes two for two and concedes he will probably miss the third by a quarter, then volunteers a fourth with a hard date attached: 2027, an AI voice on a phone call that you cannot distinguish from a human.
Around the audit sits the rest of the episode: why he thinks AI ends up 100 times bigger than the internet, why only one website in a thousand has an API and what that means for agents, why the iPhone and the MacBook are "complete rubbish for AI," why the binding constraint on humanoid robots is intelligence rather than factories, what a $36 million Meta counteroffer looks like from the losing side of it, and the fifteen years he spent nearly broke before any of this worked.
"I give zero shits about that" (0:00)
Sam opens with the number. Google Brett Adcock net worth and Fortune says $19 billion. Three years ago on this same show, Adcock described a very different balance sheet: tens of millions, almost all of it shoveled into Figure, a mortgage on his house, and the rest tied up in Archer stock that was not doing well.
His answer to the $19 billion is flat. "I don't care about that at all. Give like zero shits about that."
Sam does not accept that as the whole story, and pushes: you are a hypercompetitive guy, you have said you want to win a bunch of times, you have this Napoleon energy of wanting to conquer. Adcock reframes rather than denies. The companies are only now hitting an inflection point. They are early. If it works, it is a 100x or a 1000x from here, and that is where all of his energy goes.
Then the line that governs everything else he says for the next hour:
There is no flatline here. It is either it goes down or goes up. It is binary. Either the robots go out to scale or they do not go out to scale. So in like 5 years time, it is either going to be a very big thing or very bad.
AI is going to be 100 times bigger than the internet (2:33)
Shaan asks for the five year outlook. Sam notes that three years ago he predicted Figure would land in the $40 to $50 billion range, which it roughly has, but that Adcock still has not shipped robots at volume.
Adcock zooms out first. The claim is not incremental:
I think the AI work that we're seeing here now is going to be like a hundred times bigger than the internet. It's just working so well. Deep learning works and everything's happening faster than I would have thought.
He grounds it in his own career. Fifteen years of software and internet companies taught him a pace. Nothing on that trend line moved this fast. He returns to the same 100x number at 53:51 near the end of the episode, adding that "AI is going to eat the whole internet. It's going to eat it all up," and that the trend is large in both directions at once, physical and digital.
Hark: the AI that actually uses a computer (3:03)
Asked for a concrete thing that blew him away, he goes straight to his newest company. Hark started about a year ago around a thesis he calls AI to human symbiosis in the digital world. Figure, in his framing, is the maximum ceiling of AGI in the physical world. Hark is the same idea pointed at the screen: a person paired with an AI that eventually has its own weights, its own memories, possibly its own hardware.
The load bearing technical premise is that the pairing is worthless if the AI cannot operate a computer.
You would never hire an assistant that couldn't use a computer.
One website in a thousand has an API (4:04)
This is the specific number the whole architecture turns on. Adcock says only about one in a thousand websites has an API, and that most computer use in the world happens on the open internet through a browser. So an agent built on APIs and MCP servers can only reach the sliver of the world that publishes one.
His conclusion is that general purpose computer use has to be solved the way a robot solves the physical world: look at the screen, move the mouse, use the keyboard.
You can't rely on API or MCP. You have to figure out how to navigate like a human can.
He says his own expectation, two or three years out, is a system you talk to and hand arbitrary work: build the financial model, book the flight, order DoorDash. As of the recording Hark has released a first model as a research preview, out the week before, and he claims it topped some of the leading browser and computer use benchmarks in the world. His stated status: "It's really hard for us to find now something that we tell it to go do on the internet that it can't do."
What Hark did differently (5:06)
Shaan pushes back with the obvious objection. Everybody is doing computer use. Elon has one. ChatGPT has one. Sam adds a firsthand review: he asked ChatGPT's computer use agent to book a massage, watched the mouse move and the results scroll, and it simply did not work well. So what is different?
Adcock gives three answers, two architectural and one that he declines to detail:
Everyone else is coming at it through APIs and MCP. His example is Open Claw, which he says got great precisely because it could not use the browser, so it could never take a DoorDash order end to end, because DoorDash has no consumer API.
A virtual computer per agent. Hark spins up a fresh sandboxed machine for every agent, so nothing needs a physical MacBook and you can run as many environments as you want. On top of that the agent gets vision on the screen plus cursor and keyboard control.
The actual unlock is in post training. Pressed on whether it was an algorithmic breakthrough, he says: "It was in our post training. We have a reinforcement learning process that we think maybe nobody else in the world has done." He does not elaborate, and the hosts do not get more out of him.
Sam's aside about the fundraise is worth keeping, because it is the whole pitch in one image: the Hark deck was Adcock talking to camera for an hour, followed by a list of about a hundred engineers who had just moved from China with extraordinary backgrounds. That was the deck. The money came for the team.
The two directions AI goes (7:38)
Adcock's map of the future has exactly two branches, which he expects to converge eventually.
Physical. AI out in the world doing laundry, dishes, cooking, running supply chains, working in healthcare. The vessel is a humanoid, for a specific reason: you want one piece of hardware whose body is already capable of everything, and then you pour smart AI into it. That is Figure.
Digital. A very close AI to human pairing that goes everywhere with you, knows all your stuff, holds all your memories, has access to your accounts and systems, and actually goes and does things. His reference is Jarvis from Iron Man. It is superhuman in almost every way, always in the background, always helping. His example: you have a short layover, you miss the connection, and it already has backup plans staged before you have finished walking off the jet bridge.
His verdict on where we are: "We have really good coding agents. We have really good chat bots, but we don't have something that can go off and be my Jarvis."
To get there he says the model side needs four things beyond text chat: computer use, near perfect memory, natural back and forth speech, and vision, so the system can look at the world and understand what it is seeing.
The phone and the laptop are the wrong interface (9:39)
Then the second half of the Hark thesis, and the part that gets the most airtime:
You have AI over here and a human, and you have an old hardware system in between, like a MacBook or iPhone. They were designed 20 years ago. They're complete rubbish for AI. They're not the right interface.
He frames it as a standard upgrade cycle, the kind founders watch for. Computers and phones go away. What replaces them is always on, always listening for context, always able to reference what is happening. Most apps get abstracted away. There is probably no app store. There is an AI operating system instead, and eventually your own weights living on your own device that you own and carry.
The team is about 80 to 90 people. He notes that the person leading hardware design previously did the last several generations of iPhone, MacBook, and MacBook Pro.
The roadmap he gives on air:
First research preview of the computer using agent: shipped the week before the recording.
Hark on traditional browser, iPhone, and Android: about a month out.
Hardware: in the lab already, being tested. His description is "it's like a sci-fi movie hardware."
The billion units a year test (11:41)
Shaan asks the obvious device question. Jony Ive's shop was acquired by OpenAI, there are leaked videos of a puck, there is an earring, there was a supposed Super Bowl commercial. What do these things actually look like? Watches? Glasses?
Adcock says his own view has changed a lot in the last year, and that Hark's internal opinion is now strong and specific. The filter he uses is volume:
What sits in the middle is devices that could possibly reach a billion units a year in the world. The only things we have like that right now are computers and phones.
Everything else orbits that center. AirPods, watches, wearables. He notes they are roughly 3% of Apple's revenue and exist to support the platform, not to be the platform. Hark is going after the middle, and he is explicit that there is no gentle way to do it:
To solve that, you've got to take down the computer and the phone. There's no way around that.
Shaan tries to soften it into something familiar: so the next device is basically an AI native phone, same form factor. Adcock rejects that outright. "No, I'm not saying that at all. You're going to want to really radically rethink everything." The first hardware version in their lab is, in his words, "unlike anything I've ever seen in my whole life."
Why it is not glasses (13:13)
He is unusually blunt here, and the reasoning is concrete rather than aesthetic. On the Meta glasses:
They're probably one of the worst products I've ever bought. They're just horrible.
His complaints, in order: the glasses have no network of their own, so they piggyback on the iPhone's network; that means the app has to be open on your phone; the pairing takes forever; it does not work well. His summary is that he cannot think of a single reason to strap the thing to his head for 14 hours a day.
The forward looking version of the claim is the falsifiable part:
The end state is BCI in the brain, and we're going to have AI language devices for the next 10 years before that. That's the path, and it's not glasses. I don't even know if glasses will make our top 10 list of devices.
Zero constraints, and the human in a box (14:13)
Sam asks how the team brainstorms outside existing norms, because he genuinely cannot picture what Adcock is describing. The answer is a design method, and it is the most transferable part of the episode.
Step one: go down to the substrate and ask what actually changed. His answer is that AI is a new kind of computer, a new kind of automation.
Step two: only build around capabilities that are 10x, never 2x. His rule: "If it's one or two times better than your phone or computer, you're not going to use it. It's got to be literally 10x better." So the design question becomes: what does deep learning bring that is 10x?
His list of what is genuinely new:
It can think and use computers and systems for you, the way a person does.
It can talk to you, real time, speech to speech.
It has visual understanding. It can see.
It operates a computer at close to human speed, and over time faster, with a better success rate.
It has memory that you can write into, and it will not forget. Near perfect over time.
Put those together and you get his phrase for it:
You have a system that's almost like a human in a box that has all the same affordances a human has.
Then the thought experiment he actually designs against, which he aims directly at Sam: imagine you could carry a little human around on your shoulder everywhere you went, with a computer. Only you can see it. Only you can talk to it. It exists only to help you, for your entire life. It gets smarter every year, has perfect memory, can use computers, can talk to you, and can see.
The counterexample he uses to show how far the current stack is from that is deliberately mundane. He put a contact into his phone last week while distracted. A day later someone asked whether he had called that person and he could not remember the name, and he could not ask his phone, because his phone has no idea what happened. "It's so stupid." Then he opens DoorDash and, in his words, orders "like a monkey," pushing things around for three minutes to check out.
What he does now instead, using Hark on the drive to work: say "order me coffee," and it is done, end to end, in the background, nothing to touch. And the little human on your shoulder would go further and predict it: Brett, you want coffee today? Yeah, make it a double shot, and route it to the Hark office instead of Figure. Done.
Shaan closes the loop with the best analogy in the episode:
The phone is like a hammer. If you want the hammer to do anything functional, you have to pick up the hammer and start swinging it. The next generation is basically having a handyman next to you at all times. You just tell them, hey, can you fix that window.
Rapid prototyping (17:14)
The step after zero constraints is volume of physical iteration. Hark has a fabrication facility and a full design studio. They have designed and 3D printed, in his words, everything you could possibly think of.
Asked which failed designs were still cool, he does not really answer, and it is instructive that he does not. His explanation is that they are building many different devices covering a wide area, and the shapes are radical enough that you cannot look at one and say it resembles a known product, so the comparison does not land. The selection process is that he personally carries or wears the prototypes for weeks and months, and then they down select.
The one piece of external validation he offers: one of the biggest telecom CEOs in the world, someone who worked with Steve Jobs on the first iPhone, visited two weeks before the recording, straight from a meeting with Tim Cook. His reaction to the Hark hardware, as Adcock relays it: "This is the first time I've ever seen anybody that could possibly take out the big guys."
The constraint is intelligence, not manufacturing (18:16)
Sam offers the natural frame: across Archer, Figure, and Hark, is the hard problem always mass production? Adcock says no, flatly, and this is one of the strongest claims in the episode.
We believe now the most important constraint to really solve is building a really intelligent robot system.
His evidence is the current market. You can buy a humanoid robot from China today. He says they are "complete crap." You can joystick it around. You can press a button and it waves. It has nubs instead of hands. His analogy is the DJI drone he bought years ago: hard to set up, did not work well, flew it into a bunch of trees, and a day later wondered what he was supposed to do with it. Robots are at that stage now. You can manufacture a ton of them, and it does not help, because they are not smart.
Cars are hard. Consumer electronics are not. (19:48)
He then separates two things people conflate, and the distinction is mechanical and specific.
The reason cars are so hard is that you can't hold the part in your hand.
With phones you can always pick the part up, move it, hold it, change it. With cars you physically cannot, so you need robots handing parts to other robots to place them on the chassis, across a line of maybe 800 to a few thousand robots. If any one of them breaks, the whole line is dead. You are effectively building one giant robot whose product is a car. He notes BMW is a commercial Customer of Figure and that he has been inside their plants and a few others: "it's gnarly."
Figure sits closer to the phone end, because every part can be held in the hand. He puts it at roughly the 40% mark on that spectrum. And he pushes back on the idea that consumer electronics volume is scary at all: the world makes close to a billion phones a year, largely by hand with some automation, and no large company would say it is afraid to manufacture consumer electronics at rate when demand is that clear.
The thousandth robot (20:50)
Concrete milestone: Figure built its 1000th EVT robot for Figure 3 the week before or the week before that.
Sam asks what a thousand robots means, prototypes or Customers. Both. Figure has two large internal consumers of robots:
The engineering fleet. Every engineer needs a robot, every lab needs robots, and there is an enormous amount of testing to run.
Customers. They shipped robots to their third Customer that week.
What the robots actually do (21:20)
The current work is mostly logistics and packages, with some manufacturing logistics historically. Shaan references the live YouTube stream Figure ran, watched by hundreds of thousands and possibly millions of people, of a robot sorting packages off a conveyor belt. Adcock confirms that is a fair example of the work.
Then Shaan asks the sharpest business question of the segment: is the Customer buying this because labor is unavailable and this is cheaper today, or is it an investment in a cost curve that only works two years from now?
Adcock's answer is "no, no, no" to the second framing. The pitch comes from the Customer, not from Figure, and it is a labor crisis pitch:
Turnover above 100% per year in some areas.
A large talent shortfall, and wages going up.
No path to automating the work with conventional means.
The economics work because a robot runs multiple shifts a day, seven days a week. High uptime is the whole argument. He says Customers are getting real ROI on the contracts now, not eventually.
2.9 seconds a package, for 200 hours straight (22:51)
This is the hardest number in the episode and the one most worth remembering, because it is a falsifiable operational claim rather than a forecast.
The task on the live streamed logistics line is a real Customer use case. The Customer's requirement is 3 seconds per package, initially for 5 hours a day, 5 days a week.
Figure ran it 200 hours straight at 2.9 seconds per package.
His conclusion: "We're already at human speeds. We're already doing this here now." And the trajectory statement attached to it: "over time, it will just put billions out to these groups."
Figure 1. The only fully specified performance claim Adcock makes on air (22:51). Bars are drawn to the four numbers he states and nothing else: the Customer needs 3 seconds a package for 5 hours a day, and Figure ran 2.9 seconds a package for 200 hours continuously. The second panel is the more interesting one, because throughput parity with a human is common in demos and 200 hours of uptime is not.
Separating fact from fiction in the robot market (23:45)
Shaan lays out the enthusiast's problem honestly. Unlike most product categories, you cannot cheaply test any of this. He sees a Chinese robot for $20,000 and has no idea what it can do. He hears Elon say a million units are shipping next year. He sees 1X hold up a hand and call it the best hand ever built. He hears about a San Francisco service that sends a robot to clean your apartment today. How does a normal person tell what is real?
Adcock agrees the signal to noise ratio is out of control. "There's just so much shit out there in the market. It's really hard to tell what the hell's going on." Then he offers his own filter, which is to define the destination and work backwards.
The thing that matters, in his framing, is shipping robots autonomously, at scale, in useful work environments. Cook dinner, clean dishes, make the bed, run the supply chain end to end, work in healthcare, build a building, do logistics.
What that requires technically:
Onboard AI, running on the robot, doing genuinely autonomous work.
You cannot solve it with code. Scripted behavior does not survive the real world.
It has to run over long time horizons.
It has to move through the world and use its hands on things. In his words, it has to "move electrons around" economically.
And the explicit list of what does not count, which is most of what goes viral:
We don't care about the best robot that's doing backflips and running the fastest mile or dancing or in a parade or running outside in the woods.
Shaan lightens it with the best joke in the episode, about wanting to see a Figure robot on a smoke break at the BMW plant: "I could have been great back in high school, but I blew it. Now I'm working at a BMW factory."
Recruiting: the pitch, and why he technically assesses everyone (25:52)
Sam sets up the segment by recalling that Adcock once defended Vettery, a job recruitment marketplace, as world changing, and did it convincingly. He is good at pitching, good at raising money, good at convincing people to uproot their lives. So how does he craft the pitch?
Adcock deflects the question. The pitches are public: the Figure master plan is on the site, Archer's was up for a long time. What he says actually consumes his time is not the pitch, it is the search.
Even in the Bay Area, where it's probably the richest AI and engineering folks in the world, 90% of everybody out here is not good at their jobs.
Asked how he tells, the answer is one word: "I technically assess them." All of them. Personally.
The heuristic he uses is worth quoting in full, because it is portable to any technical hiring:
I need to know if you did the work or if you watched somebody do the work. If you've done the work, it's like a scar you carry with you. It's dug into you. You know all the details. You can talk about it freely. You don't need to think. You'll understand how to reverse engineer everything you've done and discuss it. The folks that haven't done it get one layer and they instantly blow up.
Then the number that makes the point brutal. Figure's mechanical engineer bar requires building actuators from scratch: bearings, motors, a gearbox, sensors, all in a very compact package with hard requirements.
We've been doing 10 case studies a week for six months and have not hired anybody.
Roughly 240 qualified looking candidates through a full case study, zero offers. Sam's reaction, and it is the right one: "That's insane." Adcock's reply: "It's insane."
The talent market from inside: $20 million versus $36 million (28:24)
Shaan asks what happens when you do find someone and their competing offers come from larger, more liquid companies. Adcock says the AI side is exactly like that, and that it is almost entirely driven by Meta. He thought it was a temporary spike a year ago. "They've not stopped."
Then he tells the specific story, and it is the clearest window into the market anyone gives on air.
The candidate was senior, on the AI infrastructure side, coming out of xAI after what Adcock describes as a mass exit about six months prior in which "macro hard got fully disbanded" and basically everybody left.
Hark's offer: roughly $15 to $20 million of Series A stock over 5 years. Adcock's companies use a five year vest early on and move to four later; Hark is still on five.
Hark's pitch on top of that: 10x from here gets you a few hundred million. 10x once more gets you a few billion. "I think we can do it."
Meta's offer:$36 million over 4 years, in RSUs, which the candidate treated as essentially guaranteed cash.
The candidate took Meta. Adcock's summary of the calculus from the other side of the table: maybe $200 million at Hark, maybe $20 million, versus $36 million for sure at Meta.
He says this is not an outlier. "Every candidate we speak to is making some absurd thing."
The junior band, which Shaan asks about directly, is also specified:
People in their 20s and late 20s: about $200,000 to $250,000 base.
Plus roughly $1 million a year in equity.
Total range $750,000 to $2 million a year.
And the scarcity claim underneath the whole market:
My rough back of the envelope is probably like 20 to 30 people in California know how to build really good AI models.
Figure 2. Every compensation number Adcock states between 28:24 and 31:28, normalized to dollars per year so the offers are comparable. The Hark and Meta bars are the same candidate, the same week. The gap is not the interesting part: the risk profile is. Meta's is RSUs the candidate read as guaranteed cash. Hark's is Series A stock in a company that is one year old, which Adcock argued could be worth a few hundred million on a 10x and a few billion on a second 10x. The candidate took the certain money.
How Zuck is buying his way into AI (32:03)
Asked what he thinks of Meta's strategy, even if he hates it, Adcock's answer surprises both hosts: "I really like it."
His reasoning is scarcity. The people who genuinely understand language model pre training, mid training and especially pre training, plus the infrastructure around supercomputing, data, and evals, are vanishingly rare. Knowing which recipes transformers actually respond well to is not something you can hire for on a resume. Hence the 20 to 30 people in California figure.
Sam asks whether he was being sarcastic. He was not.
I think it was really smart and I would have done the same thing if I was Mark. I would have bought my way into the race.
Then, immediately, the correction that makes it interesting: he would not do it himself. Not because it is wrong, but because he wants to understand the thing from first order and find people who care deeply about it rather than hiring mercenaries.
And he is unsparing about what Meta actually bought:
Nobody wants to go to Meta. They're going there because they're getting paid a guaranteed RSU package by sitting around.
The structural argument underneath is the one worth arguing with, because it cuts against the entire spending war:
You don't need a thousand people or 500 or 300 to design AI models. You need a really good team of 20 or 30 or 40 people. And you can get there without doing this.
His verdict is split cleanly. Execution: hats off, the recruiting is working, the packages are structured well, it is paying off. Products: jury is still out. His stated problem with those groups is that they have historically not been able to do genuinely new things. He grants that Meta will go down as one of the greatest acquirers of all time on the strength of Instagram and WhatsApp, and then says the Ray-Ban work "is not great work."
He extends the criticism to the whole market, and it lands as the thesis behind why he is building Hark at all:
Every week there's like five or 10 junk AI slop startups or things coming out. They're just not very good. The stuff in coding is probably really excellent right now, but everything beyond that is just kind of not great.
The demo aside: he says they are running Hark on the podcast at that moment, and that "it's so much better than anything I use today." Both hosts immediately ask for access. He says yes, noting the research preview currently consists of "like 500 PhDs and then me and Sam."
The audit: grading the January 1st predictions (33:28)
This is the segment the title is selling. On January 1st, 2026, Adcock published predictions for the year. Sam reads them back with roughly four months left in the year and asks for a verdict on each: on track, off track, or done.
Prediction 1: humanoids working unsupervised in a house they have never seen
The prediction, as written and read verbatim by Sam:
Humanoid robots will perform unsupervised multi day tasks in homes they've never seen before, driven entirely by neural networks. Long time horizons, going straight from pixels to torques.
Verdict: on track. With four months left. He says he sees the progress daily.
He then separates the easy half from the hard half. Pixels to torques is already solved and he laughs about it being in the prediction at all: they take camera feeds in and output motor commands, telling joints where to move to put the hand in the right place. That part ships today.
The hard part is generalization to an unseen home, and he is precise about the failure mode. Figure can fold laundry. Move it to a new place with different lighting, a different table height, different types of laundry, situations it has never seen, and:
The model is out of distribution. It doesn't know what to do. It's like if you removed all the pyramid data from the pre training of an LLM. It wouldn't know how to talk about pyramids.
And the resulting bottleneck is a data collection problem, not an architecture problem:
We just don't have enough of that data out there. It's not on the internet. So you have to go out and collect it.
The open research question he says he is actually working on, three to four hours a day, every single day, seven days a week: how much of that data do you have to sample in the world before the model can walk into your house and fold clothes on command.
Why laundry, when warehouses pay (35:30)
Shaan pushes on the strategy with a genuinely good question. There are already 20 million warehouse workers in the world. That is a huge, well defined, paying market with none of the sensitivities of somebody's living room. Why care about the home at all today?
Adcock's answer is that his strategy changed, and he says so plainly. The original plan was the standard one: do the commercial side to fund the home long term. Commercial charges more, has lower variability, sits inside a controlled work site, runs 24/7, and is much simpler.
What I've learned now is that the home is super solvable today.
So the choice is not economic, it is about who wants to solve what:
We could not work on that problem and just sit here and work in a warehouse, but me and none of my guys want to solve that problem. We want to solve a robot that can go into any environment, just through language, and do work. We want to be the first to do that.
Then the claim that should be flagged in neon, because it is the biggest single number he says all episode:
You can probably do that with a 100 robots and a 50 person team. So that company overnight would be a trillion dollar market cap.
Sam: "That sounds good. Do that." Adcock: "We're doing that. We're going to solve that. I think we'll be the first. We call it solving general robotics."
Shaan asks about I, Robot, and whether the humans get attacked. Adcock says do not worry about it. Asked whether a human can still win a fight against a Figure robot right now, he says yes.
Prediction 2: past text, into voice with persistent memory
As read:
Daily AI usage will shift. People will move beyond text to highly multimodal. Voice agents with persistent memory will become common, which will push AI closer to the synthetic human intelligence we've imagined in sci-fi.
Verdict: on track. His evidence is his own roadmap: Hark ships the first version of exactly this in a month, and it improves from there.
The hosts corroborate the behavior shift from the user side. Sam says he barely types at all any more. Shaan wonders whether the labs have ever published data on voice share. Then the funniest true observation in the episode: open plan startup offices are now the wrong shape for how people work, because engineers are sitting there whispering to their computers with microphones, in hushed tones. Shaan says he asked Claude "why am I so indecisive" out loud the previous night and his wife heard the entire thing. "She can hear everything I'm talking to Claude about now."
Adcock's own assessment of the state of the art is much harsher than the prediction implies:
Even speech still sucks. It's still not great. You have to go there, you have to turn it on, it doesn't really remember what you just talked to it about. It can't do tool calling and computer use very well. It's limited to a session.
Prediction 3, made live: the 2027 speech Turing test
Immediately after grading the voice prediction, Adcock volunteers a new one with a hard date, and it is the most falsifiable thing he says all episode:
I don't know if we'll hit it this year, but certainly in 2027 you will hit a full human Turing test with speech. You'll be able to take a phone call from an AI system on your phone and it will be able to fool you guys. I'll be able to have a human call you and a robot call you and I don't think you'll be able to tell the difference. That's a 2027 event I feel pretty strong about.
Note the structure. It has a date, a test protocol (a paired call, human versus AI, the listener must identify which), and a stated confidence level. Nothing else in the episode is specified that tightly.
Prediction 4: weapon detection at a 20 foot standoff, in a K12 school
As read:
Over the past 10 years, school shootings have increased by 10x. In 2026, the first full scanning system capable of detecting weapons from a 20 foot standoff will be built and beta tested in a K12 school.
Verdict: partial miss, by about a quarter. This is the only one he grades down, and he does it without hedging.
Cover starts building the full scale system in October. He thinks they bring it up before the end of the year. He does not think it will be inside a K through 12 school by then. "So we might miss this one by a quarter."
The reason is the most interesting engineering story in the episode, and it is a story about semiconductors, not about AI:
Cover would already be in market. Adcock pivoted the entire technology stack about a year ago.
He found a way to do the whole thing in silicon and chips, which cut the price by roughly 90%, made it far more scalable, and made it work better.
The cost was schedule. Designing custom chips and getting them fabricated took about a year.
The chips arrived a couple of months before the recording. They are testing them. "They're awesome."
Making more of them carries another six month lead time.
These are custom designed Cover chips that, in his words, nobody has really designed before, made by a special fabricator in Europe.
Organizationally: Adcock is CEO, with a chief engineer out of JPL and NASA running it. He describes it as almost pure engineering with very little business surface. There is supply chain work, but the question is simply whether you can build a system that detects weapons. It is partly a hardware problem, partly an AI problem, and entirely a deep tech problem.
The bookkeeping note on "four predictions"
Worth being precise, since the title sells a number. Sam asks at 38:04 for "the third and fourth" from the January list, Adcock gives the school scanning one, and the conversation moves on to how he structures his life. The fourth item from the published January list is never read out on air. What the episode delivers instead is three audited predictions from the list plus one new prediction volunteered live with a 2027 date on it. That is the honest accounting, and the four in the table below are the four the episode actually puts on the record.
The four predictions
Prediction
As Brett Adcock states it
Timeline
The reasoning, in one line
What would falsify it
His grade on air
1. Humanoids work unsupervised in an unseen home
Humanoid robots perform unsupervised multi day tasks in homes they have never seen before, driven entirely by neural networks, long time horizons, straight from pixels to torques (33:28)
End of 2026
Pixels to torques already works; the only gap is enough real world data to stop the model going out of distribution in a strange room
End of 2026 arrives with no Figure robot completing a multi day task in a house it was not trained in, or the demo still requiring teleoperation or per home setup
On track
2. Daily AI use moves past text
Daily AI usage shifts beyond text to highly multimodal; voice agents with persistent memory become common, pushing AI toward the synthetic human intelligence of sci-fi (37:02)
End of 2026
Speech today is session bound, forgets, and cannot do tool calling or computer use; Hark ships a version that fixes all three in about a month
Voice remains a session bound novelty with no persistent memory and no tool use through the end of 2026, and typed chat stays the dominant modality
On track
3. A speech Turing test that fools you
A full human Turing test with speech; a human calls you and an AI calls you, and you cannot tell which is which (38:04)
2027, volunteered live rather than from the January list
Multimodal expressive models plus real time speech to speech plus memory converge to close the last gap in a phone call
Run the paired call in 2027 and listeners still identify the AI reliably above chance
"A 2027 event I feel pretty strong about"
4. A 20 foot standoff weapon scanner in a school
The first full scanning system capable of detecting weapons from a 20 foot standoff is built and beta tested in a K12 school (38:34)
Calendar year 2026
Cover rebuilt the detector in custom silicon, cutting cost roughly 90%, at the price of about a year of fabrication lead time
The system is not standing and detecting by December, or the standoff distance comes in materially under 20 feet
Partial miss, by about a quarter
Figure 3. The four predictions positioned by deadline and confidence. The horizontal axis is the deadline each prediction carries in the episode, which is stated for all four. The vertical axis is not measured: it is ranked from Adcock's own words during the audit, so "on track" with a product shipping in a month sits above "on track" with an open research problem, which sits above "I feel pretty strong about," which sits above "we might miss this one by a quarter." The dashed arrow is his own slip estimate on Cover, one quarter past the deadline, not an outside forecast.
Three buckets, and the one he deleted (40:34)
Shaan asks the price question: you are firing on all cylinders professionally, so what is the trade off?
Adcock says he hit this about five years ago, with young kids and multiple companies running. His model of his own life has exactly three pockets:
Work, which he cares deeply about.
Family, three kids, all young.
The other stuff. A friend is in town. The annual golf trip. A bachelor party. A wedding in Europe.
His conclusion was that he could not be A+ at all three. So he did not try to balance them. He deleted the third one.
The example he gives is not softened. His freshman year college roommate was in the Bay Area for ten days and asked to get a coffee. Adcock told him the truth: I have no time. The friend offered to come to him, anywhere, any time. Same answer.
Every minute I'm away from one of these two is a minute I'm away from my family or work. And there's a limited amount of time I can put in both those buckets.
Show us your home screen (42:05)
Sam asks him to hold up his phone, and the answer is more interesting than the usual app tour because of what has been replaced.
The lock screen is wall to wall notifications, mostly Slacks and texts.
He does use Slack. He cannot get through it during the day. Hark reads it for him, and then texts him when something is important, with a link.
Setup is laptop plus phone, laptop a lot.
All AI work runs through Hark, end to end. Tracking engineering projects, recruiting, all of it. It sits in his email and his Slack.
The to do list lives in Hark. Before Hark it was a Google Doc called "replanning" that he updated every Sunday for the coming week.
Health (43:05)
He has access to a concierge medicine setup: quarterly blood panels, full body scans, cardiac CT. What impressed him is the sheer volume and thoroughness of the data that comes back. The specifics he names:
A CT scan of the heart for about $100, which he thinks can basically prevent heart attacks.
A full body MRI for early cancer detection.
Broad blood work to catch anomalies you can go fix.
There are maybe a dozen such tests.
Sam does not let it stand. The fixes those tests point to are the ones Adcock is unwilling to do: get up, walk, exercise, all of which fall in the bucket he deleted. Adcock concedes it directly. He has not had enough time to exercise enough. He does eat well now.
Something's got to give. I can't sit here all day eating. I've got to go work. I want to go crush these businesses.
Rock bottom, three times (44:07)
In the show's prep document Adcock wrote that he went all in on his first three startups and "pretty much hit rock bottom every year." Sam asks him to define rock bottom and describe the method for getting out.
The definition is financial and it lasted fifteen years. "I was basically always running out of money."
2012 to 2017The Vettery years. Multiple pivots early. Debt. Nothing working.
2015Raises a $500,000 convertible note. Takes out a personal loan of $50,000 to $100,000. Pays himself no salary. Living in New York City, in his words "so broke, I was in the negative." The round "did not look great."
2015, six months laterLaunches the Vettery marketplace. It takes off immediately.
2016 to 2017Sells Vettery for $110 million, a roughly 12x to the venture investors.
ThenStarts Archer. Everyone he pitches says no. "What are you doing? We're not going to fund this. What are you talking about?"
Archer yearsBuys a house, puts the rest of the money into Archer, and is locked up on the stock. The stock falls from about $10 to about $2, "literally a falling knife," while he is trying to fund the next company with it.
Founding FigureFunds Figure with unlocking Archer stock because there is no other cash, and takes a second mortgage on his house to keep it going.
2026Fortune puts his net worth at $19 billion. Figure is valued somewhere in the $30 to $50 billion range. He says he gives zero shits about the number.
The method: punch list, day to day, and the half mile trick
Asked for the inner monologue, he does not dress it up. "This really sucks. Super painful."
The method is deliberately small:
There's only one way out and that's through. So you need to build a punch list and you need to get to day to day. You've got to go day to day. You can't go week to week. Two days, look at Friday. You've got to get to the next day. Pile through it.
The story he credits for it comes from training for an ultramarathon, a distance he says he hates. He read about a runner whose whole method was to pick something in the distance, a hundred feet or half a mile, and tell yourself that once you get there, then you will consider quitting. You get there. You decide maybe you have a little more. You pick something two hundred yards out. You will consider quitting when you reach that. You have 49 miles to go and it does not matter.
The survivorship number
Shaan asks how you fight the monologue when everyone tells you that you are stupid and wrong and should just go do software. Adcock's answer is conviction earned by doing the work: "I know I'm right because I've done the work. I understand it. I'm on the floor."
Shaan presses: the odds are still against you.
That's the game. You sign up and 95% of everybody around you will fail.
Then the concrete version. Vettery started at the NYU incubator in Soho, in a cohort of about 50 companies. Five years later, by his count, he and one other founder were the only two people who made more than zero dollars. Forty eight companies went to zero.
If you're around this game long enough, everybody dies, and that's everywhere. It's been like that for 20 years now. You see all the TechCrunch stuff and things on X about people raising money, and over time that all just fades away. It's really brutal.
"Doing hard things is easier than doing easier things" (48:13)
The hosts read back one of his stated philosophies and ask him to defend it. His argument has five parts, and it is the closest thing in the episode to a general theory of what to work on.
Less competition. Everybody is trying to do easy things, so the easy space is crowded and the hard space is not.
Bigger addressable market. A hard thing usually implies a large TAM and a large exit if it works.
Better people want in. The overachievers want to work on hard problems, so the hiring pool is self selecting.
Investors prefer the shape. Hard things have a binary payoff, which is exactly what a portfolio wants, because one 100x return carries the fund.
The difficulty curve is nonlinear, and this is the actual insight. Hard things are not 100 times harder. They are usually two, three, four, maybe five times harder, while the payoff can be 100 times better.
Humanoids versus robot dogs
The worked example. Building a quadruped, a four legged robot dog, versus building a humanoid:
Humanoids are roughly three times harder, maybe four.
There is, in his view, no real market for robot dogs. It is a niche. He does not think there is a real business in it, and he does not know anyone who wants to spend a lot of time on it.
So the humanoid is three times harder for what he estimates is a million times or a billion times higher return, for investors, for the people who work there and hold stock, and for everyone else.
What economic value can a robot dog bring at scale?
The absolutes, and the Open Claw jab
He extends the same argument to AI wrappers:
Look at all the AI slop Open Claw harnesses out there today. It's all crap. It's all not good. I don't think any of them will make it long term. You might have some consolidation here and there for acquihires, but that's going to go all the way.
Sam calls him on the style: "Dude, you talk in so many absolutes. Has that not gotten you in trouble ever?" Adcock does not back off. "Mark my words. Have you guys even used Open Claw since then?" Neither host has. Shaan never trusted it enough to set it up. Sam used to and does not any more: "It's not very good. The wave's over."
The point he lands on is about founder time allocation, not about any particular tool:
The most important thing you can do as a founder is to think through what you're actually going to go do, because you're going to spend the next 10 or 15 years doing it. It's a probability weighted decision of potential outcomes.
The pushback: not everyone is playing a binary game
Shaan makes the best counterargument in the episode and does not let it go. Adcock is playing a specific game where the outcome is trillion dollar company or bankruptcy. Most business is not binary. For most people, building a really cool $10 million a year business is a massive home run.
Adcock's answer is a question rather than a rebuttal: if you built a cool $5 or $10 million business, ran it for thirty years, never tried anything else, and looked back at 70 or 80, would you go back and take a bigger swing?
He then turns it on himself, and this is the most self critical passage in the hour. Vettery was that business.
Vettery is my bridge. I sat inside of Vettery for like seven years. We literally built a marketing automation tool for us internally. And then a year later I was like, oh man, look at this, it's Outreach.io, and it was a billion dollar company. We built that internally a year or two prior.
The lesson he draws is not "we should have shipped it," it is broader: what you choose to spend time on is the critical decision, and he watched value he had already built internally get captured by other people because it was not what he was pointed at.
He also concedes Shaan's point at the end, genuinely. He believes startups are binary even at the $10 million level, because for every founder who gets there another 90% did not make it. "Kudos to guys getting five or 10 million in a business. That's hard. Especially doing that with a little bit of capital or no capital."
One piece of hardware in a human form that can physically do everything, with smart AI poured into it. Laundry, dishes, cooking, supply chain, healthcare. "Solving general robotics" means walking into any environment and doing work from a language instruction alone.
1000th EVT robot of Figure 3 built. Third Customer shipped that week. BMW is a commercial Customer. Live streamed a logistics line at 2.9 seconds a package for 200 hours.
Generalization, not manufacturing. In an unseen home the model goes out of distribution, and the missing training data does not exist on the internet, so it has to be collected. Binary outcome on a five year horizon.
An AI paired to a human in the digital world, eventually with its own weights and memories on hardware you own. It has to drive a computer with cursor and keyboard, because only about one website in a thousand has an API. Then it needs a new device, because the phone and the laptop are "complete rubbish for AI."
About a year old, 80 to 90 people. First computer using model out as a research preview. Browser, iPhone, and Android in about a month. Hardware in the lab. Raised in the billions largely on the strength of the team.
Two hard problems at once, frontier models and a new consumer device, against Apple and OpenAI. His own filter says a device has to reach a billion units a year to matter, which is a bar essentially nothing has cleared.
Full scanning that detects a weapon from a 20 foot standoff, aimed at schools. Partly hardware, partly AI, almost entirely engineering. Run by a chief engineer out of JPL and NASA.
Whole technology stack pivoted about a year ago into custom silicon, roughly 90% cheaper. Chips arrived a couple of months ago and test well. Full scale build starts in October.
Semiconductor lead times, and nothing else. A year to fabricate, another six months to make more, from a special fabricator in Europe. The schedule slip is the prediction miss.
A recruiting marketplace. He calls it "my bridge."
Sold for $110 million, roughly 12x to investors, after five years of debt and pivots.
The one he criticizes himself over. Seven years inside it, during which the team built a marketing automation tool internally that turned into a billion dollar category owned by someone else.
What he would build if he were starting now: energy (52:19)
Sam asks what other founders he has met are doing that shows the future. Adcock does not name a company. He names a sector, and it is the one thing all episode that is not his own business.
I think we have an energy problem. Not an energy consumption problem, a generation problem. How do we generate more energy as a species?
He frames it as a secular trend worth riding, and he grounds it in the literature correlating energy availability with human standards of living. The open question is which generation technology wins: solar, wind, nuclear, and inside that, fusion and fission and the rest. He says it will take a long time and that it needs great entrepreneurs working on it.
Then he returns to the 100x claim, which is where the episode's forecast really sits:
AI is going to dominate a lot of stuff in the next 10 or 20 years. We all lived through the internet. I think it's going to be 100 times bigger than the internet. AI is going to eat the whole internet. It's going to eat it all up. Extremely large trend, both physically and digitally.
Nothing on the market impresses him (54:22)
Asked for products outside the mainstream that show what he is describing, he says it is genuinely unclear who wins. The space is foggy. There have been early wins and early breakouts but no real breakout, and there is a next leg coming that he thinks resolves in the next year or two.
His personal verdict on the current device market, from someone who has bought all of it: "I've used every AI device out there. Haven't been super thrilled. I haven't been like, man, this is a crazy great product."
The only product either side of the table praises is Shaan's, and it is small: Wispr Flow, which he calls Whisper Flow on air. He says it has meaningfully changed how he communicates and is his standout of the last six months.
Who inspires Brett (55:23)
The pattern he names is not achievement, it is total dedication to a craft.
Michael Jordan. He points at the documentary. "He just wanted to be the best in the world at this." He thinks startups have the same thing.
Steve Jobs. Never met him. "From the stories I've heard, the guy was an unbelievable operator and product led founder."
Jeff Bezos. An investor in Figure, and someone Adcock has come to know well. He has been in the office many times and is a sounding board.
Jensen Huang. Was at Figure the week before the recording. Adcock says they are fairly close, calls him an unbelievable operator, and notes his unique hands on way of managing Nvidia over thirty years.
The Bezos advice
The most specific piece of counsel in the episode, relayed from a previous visit:
You're at a really interesting period, because you figured out how to do this somehow. In the next year or two, you're either going to figure out how to break through and really get this working in a bigger way, or you won't. This is game time for you now. You've got to get wired in and figure out how to break out and make this thing work and scale it. I don't know how you got here and I don't know why you got here, but you're here, and your next big push is going to make or break it.
Adcock thinks he is largely right.
The knee
The episode ends where it should, with Shaan's memory of visiting Figure four years earlier. There were five engineers. There was a knee. That was it. Adcock introduced the team: this guy just finished building the Tesla Model X or the Cybertruck, this guy did that amazing thing, this guy cured cancer. And everyone stood around watching the knee move and the ankle do dorsiflexion, and it was the coolest thing in the world.
And now we have AI working on a humanoid robot. We're taking in cameras, it's doing inference onboard, it's outputting all the joints. It's unbelievable. And it's crazy. It works. The next leg up is just making that work at higher scale.
Key takeaways
The guest is Brett Adcock, founder of Figure. The title's "$39B tech founder" is Figure's valuation, which Adcock himself puts loosely at "30 something, 40 something, 50 something billion." Fortune has his personal net worth at $19 billion and he says he does not care.
The audit is the point. Two of the January 1st predictions get graded on track, one gets graded a partial miss by a quarter, and a fourth gets volunteered live with a 2027 date. Very few forecasters submit to this in public.
One website in a thousand has an API. That single number is the entire architectural argument for building agents that use a screen, a cursor, and a keyboard rather than API and MCP integrations.
The claimed unlock is post training. Adcock says Hark's advantage is a reinforcement learning process "maybe nobody else in the world has done," plus a fresh virtual computer spun up per agent. He gives no further detail.
Intelligence is the constraint, not factories. You can buy humanoid robots today and they are useless. The world already builds close to a billion phones a year. Cars are hard because you cannot hold the part in your hand; humanoids are closer to phones on that spectrum.
2.9 seconds a package for 200 hours straight. Against a Customer spec of 3 seconds a package for 5 hours a day. Throughput parity plus uptime is the claim that matters.
The home is now solvable, so they are doing the home first. A reversal of the original plan to fund the home with commercial revenue. He claims general robotics needs about 100 robots and a 50 person team, and that whoever does it first is worth a trillion dollars overnight.
The out of distribution problem is a data collection problem. Fold laundry in a new room with different light and a different table and the model has never seen it. The needed data is not on the internet.
The AI talent market, with real numbers. A senior Hark offer of $15 to $20 million over five years lost to $36 million over four years at Meta. Juniors clear $750,000 to $2 million a year. He estimates 20 to 30 people in California can actually build good models.
Meta's strategy is smart and he would not copy it. He respects buying into the race, but argues you need 20 to 40 excellent people, not 300 to 1000 paid ones, and that mercenaries do not do new things.
Design against 10x, not 2x. Strip constraints, ask what deep learning genuinely makes ten times better, and build only that. The mental model is a human in a box: sees, talks, remembers perfectly, operates computers at human speed.
Glasses are not the answer, and he is rude about it. No independent network, requires the phone app open, slow pairing. The end state is BCI, with about ten years of AI language devices in between.
Hard things are three to five times harder for a hundred to a million times the payoff. The difficulty curve is nonlinear and most people misprice it. Humanoids versus robot dogs is the worked example.
Fifteen years of near bankruptcy preceded any of this. Personal loans, no salary, a $500,000 convertible note, a stock that fell from $10 to $2, and a second mortgage to fund Figure. Of about 50 companies in his NYU incubator cohort, two made more than zero dollars.
He deleted a third of his life on purpose. Work and family stay. Friends, weddings, golf trips, and the college roommate in town for ten days do not.
Chapters
0:00:00 Intro
0:03:29 Hark, a human in a box
0:08:00 zero constraints
0:17:32 rapid prototyping
0:21:38 What are the robots doing?
0:23:45 what about the hype is real?
0:27:54 finding the best people
0:30:14 20M dollar salaries for engineers
0:32:03 How Zuck is buying his way into AI
0:35:44 Auditing Brett's 2026 predictions
0:42:33 reducing your buckets
0:44:00 show us your home screen
0:46:17 hitting rock bottom 3x
0:57:22 who inspires Brett
Notable quotes
There is no flatline here. It is either it goes down or goes up. It is binary. Either the robots go out to scale or they do not.
Brett Adcock, 2:03
I think the AI work that we're seeing here now is going to be like a hundred times bigger than the internet.
Brett Adcock, 3:03
You would never hire an assistant that couldn't use a computer.
Brett Adcock, 4:04, on why agents must drive a browser
You can't rely on API or MCP. You have to figure out how to navigate like a human can.
Brett Adcock, 4:35
It was in our post training. We have a reinforcement learning process that we think maybe nobody else in the world has done.
Brett Adcock, 5:36, the only thing he says about how Hark works
We have really good coding agents. We have really good chat bots, but we don't have something that can go off and be my Jarvis.
Brett Adcock, 9:09
You have AI over here and a human, and you have an old hardware system in between, like a MacBook or iPhone. They were designed 20 years ago. They're complete rubbish for AI.
Brett Adcock, 9:39
The metaglasses are probably one of the worst products I've ever bought. I can't think of any reason why I would need this thing strapped to my head for 14 hours a day.
Brett Adcock, 13:13
If it's one or two times better than your phone or computer, you're not going to use it. It's got to be literally 10x better.
Brett Adcock, 14:43
You have a system that's almost like a human in a box that has all the same affordances a human has.
Brett Adcock, 15:14
The phone is like a hammer. If you want the hammer to do anything functional, you have to pick up the hammer and start swinging it. The next generation is basically having a handyman next to you at all times.
Shaan Puri, 17:14
The reason cars are so hard is that you can't hold the part in your hand.
Brett Adcock, 20:19
We did that 200 hours straight at 2.9 seconds a package. We're already at human speeds.
Brett Adcock, 23:21
Even in the Bay Area, where it's probably the richest AI and engineering folks in the world, 90% of everybody out here is not good at their jobs.
Brett Adcock, 26:22
If you've done the work, it's like a scar you carry with you. It's dug into you.
Brett Adcock, 26:52
We've been doing 10 case studies a week for six months and have not hired anybody.
Brett Adcock, 27:22
My rough back of the envelope is probably like 20 to 30 people in California know how to build really good AI models.
Brett Adcock, 30:27
I would have done the same thing if I was Mark. I would have bought my way into the race.
Brett Adcock, 31:28
You don't need a thousand people or 500 or 300 to design AI models. You need a really good team of 20 or 30 or 40 people.
Brett Adcock, 31:58
The model is out of distribution. It doesn't know what to do. It's like if you removed all the pyramid data from the pre training of an LLM.
Brett Adcock, 35:00
You can probably do that with a 100 robots and a 50 person team. So that company overnight would be a trillion dollar market cap.
Brett Adcock, 36:31
Certainly in 2027 you will hit a full human Turing test with speech. That's a 2027 event I feel pretty strong about.
Brett Adcock, 38:04
So we might miss this one by a quarter.
Brett Adcock, 38:34, grading his own school scanner prediction
Every minute I'm away from one of these two is a minute I'm away from my family or work.
Brett Adcock, 41:35
There's only one way out and that's through. So you need to build a punch list and you need to get to day to day.
Brett Adcock, 45:40
48 companies went to zero. If you're around this game for long enough, everybody dies.
Brett Adcock, 47:12, on his NYU incubator cohort
A lot of the hard things are not 10 or 100 times harder. Sometimes they're two or three or four times harder. But you might have a hundred times better payoff.
Brett Adcock, 48:43
Look at all the AI slop Open Claw harnesses out there today. It's all crap.
Brett Adcock, 49:44
Dude, you talk in so many absolutes. Has that not gotten you in trouble ever?
Sam Parr, 50:14
Vettery is my bridge. We literally built a marketing automation tool for us internally, and then a year later I was like, oh man, it's Outreach.io, and it was a billion dollar company.
Brett Adcock, 51:48
I think we have an energy problem. Not an energy consumption problem, a generation problem.
Brett Adcock, 53:20
AI is going to eat the whole internet. It's going to eat it all up.
Brett Adcock, 53:51
Your next big push is going to make or break it.
Jeff Bezos, relayed by Brett Adcock at 56:24
Figure, humanoid robots. The Figure master plan he says is published, and the package sorting live stream with hundreds of thousands of views, both live on the Figure site and channel
Hark, the AI lab building the computer using agent and the post phone device
An honest weighing, at the end where it belongs, of what Adcock puts on the record against what can be checked from outside.
What is genuinely strong here. He submits to a public audit of dated predictions and grades one of them a miss. That is rare and it should be credited. He also gives real operational numbers rather than vibes: 2.9 seconds a package, 200 hours, 1000 EVT robots, three Customers, 10 case studies a week for six months with zero hires, a $36 million counteroffer that beat him. Numbers like these are checkable later, which is the whole difference between a forecast and a mood. And his diagnosis of the API problem is correct in the plain sense: most of the web has no programmatic interface, so an agent that can only call APIs is confined to a small, well funded corner of the internet.
Where the reader should hold the claims loosely. Almost every capability claim is first party. Hark's benchmark result ("top on some of the leading browser and computer use benchmarks") is stated without naming the benchmark, the score, or the competitors. The reinforcement learning advantage is asserted and then explicitly withheld. The 200 hour logistics run is Figure's own instrumentation on Figure's own line. None of that means it is untrue, and all of it means an outside observer cannot yet verify it.
The prediction most likely to be judged clearly. Number four, the school scanner, because it has a physical artifact and a date, and because he has already told you the failure mode: silicon lead times, not AI. He also gave the schedule, which means the October build start is a checkpoint anyone can look for.
The prediction most likely to be argued about. Number one. "Unsupervised multi day tasks in homes they have never seen" has a lot of joints in it, and each one is a place where a demo can satisfy the letter and miss the spirit. Multi day is doing heavy lifting. So is unsupervised. So is never seen before. A single staged home visit with a supervising engineer offscreen would probably be claimed as a hit and would not be one.
The claim that deserves the most scrutiny. The trillion dollar line: solve general robotics with 100 robots and a 50 person team, and the company is worth a trillion overnight. Every part of that is contestable. Nothing about a capability demonstration produces a trillion dollars of enterprise value without deployment, unit economics, supply chain, safety approval, and Customers, none of which he addresses. It is the one place in the episode where his usual discipline about constraints goes quiet.
The internal tension worth noticing. He argues that the AI talent market is irrational, that Meta bought mercenaries, and that you only need 20 to 40 excellent people. He also argues that the 20 to 30 people in California who can build good models are the scarcest resource in the industry. Both can be true, but together they explain the price he says is absurd. If the input is that scarce and that decisive, $36 million over four years is not obviously an overpay, and his own losing bid at $15 to $20 million is the evidence.
On the hard things thesis. The strongest form of it, that difficulty scales roughly three to five times while payoff can scale a hundred times or more, is a real and underused insight. The weakest form of it is survivorship. He is the founder from a cohort of about 50 who cleared zero, telling you that the odds are brutal and to take the bigger swing anyway. He is aware of this, which is why the Vettery self criticism is the most useful part of the episode: even his own bridge company, by his account, watched a billion dollar product walk out of its own building.
One bookkeeping correction to the title. The episode audits three predictions from the January 1st list and adds a fourth live. The list's actual fourth item is asked for at 38:04 and never answered. The four collected in the table above are the four the episode puts on the record, which is not quite the same thing as the four he published in January.
Full transcript
[00:00:00] I think if you Google Brett Adcock net worth according to Fortune, you're worth $19 billion. So that's like a pretty good swing. How does that make you feel? >> I don't care about that at all. Give like zero shits about that. >> Okay. So you uh Brett Adcock, the the short of it is that you were raised in a rural area of Illinois. You started a company called Veterary, which you sold for over $100 million. Then you took a company public called Archer which is
[00:00:30] like unmanned uh flying planes I guess helicopters. And then now you have a company called Figer which is worth I don't know how much 40ome 30 something 50 something billion dollars. You have another thing called cover which stops uh or aims to stop school shootings. And then now you have a new thing called HARK which you've raised money at in the billions of dollars. And you seem worn out. >> Great. You look like busy man. So, you've been on this is your third time on. I think you I think you've been on one time each year the last three years. You said uh you were telling a
[00:01:01] story about how I think it was right when Figure started. You basically said like I had I was worth I don't know how much tens of millions of dollars. I put almost all of it into Figure to get started and at one point you were like I have a mortgage on my house and the rest of my money is in Figure and some of the money is in Archer and that's not doing so great right now. And since then, I think if you Google Brett Adcock net worth, according to Fortune, you're worth $19 billion. So that's like a pretty good swing. How does that make you feel? >> I don't care about that at all. Give
[00:01:32] like zero shits about that. >> You're a super competitive guy. I think you said something like I just want to You said like win a bunch of times. Last time we hung out, it was like I want to win for these reasons. I I'm I I'm very competitive. I I want to kick ass. I think that like you definitely have to care about this a little bit and you actually have to I think you care a lot about figure being the biggest company in the world. You talk about like you definitely have this like Napoleon energy of like I want to be the best. I want to conquer. >> I I think any way I would characterize is like we're just like we're just now
[00:02:03] like these companies of mine are just now hitting inlection point and they're really early like they can be like really big. So if it works this will like 100x,000x from here. So most of my energy is like how do I make sure that works? There is no flatline here. It's either like it goes down or goes up, right? Either like it's binary. Either the robots go out of scale or they don't go out of scale. So in like 5 years time, it's either going to be a very big thing or very bad. And so all my energy is going into making this like thousand or a millionx from
[00:02:33] where we're at here. And so it's it's like the pressure is on to like really just deliver. >> Where where are you now? What's the outlook now for the next 5 years then? I think last time you were on 3 years ago, we said that I think I said it. Uh I was like you'll probably be in the $40 to $50 million valuation range, which I think you are now. But in terms of like you you're still lacking output of robots, like you still need we still need that to come. When where are you going to be in five years? What's your prediction? >> I think at a high level, I think the AI
[00:03:03] work that we're seeing here now is going to be so much it's going to be like a hundred times bigger than the internet. It's just like everything is just so it's just working so well. like the system is working well like deep learning works and everything's happening faster than I would have think and my like you know having done like 15 years of like software and internet like it was just like nothing was happening faster on a trend line here it's happening like that in AI >> can you give an example of something that has happened that's blown you away >> we started at so harkc I have a new AI lab called hark about a year ago I was
[00:03:34] like very interested in this idea of like kind of building this AI to human symbiosis digitally it's like Um, figure is going to be like I think figure is going to be like the max ceiling of AGI of like being able to put that out and then there's going to be a version of this in the digital world. It's going to be like a human is going to have this like AI pairing. It's going to have like ultimately maybe your own AI weights, your own memories, maybe your own hardware. It's going like really close. And fundamental to that thesis was like you got to figure out how to get AI to use computers general purpose. You would never hire an assistant that couldn't use a computer. So you got to be able to
[00:04:04] like give things out to it that can like do everything you can do. financial models, book flights, like order Door Dash, whatever you need to do, you need to be able to do it all autonomously, but only one in a thousand websites have APIs. So, in most, you know, glo like most computer use globally is on the on the internet and browser. My my main inclination within two or three years, you'd have a system that you'd be able to talk to and say, "Go do this or do that." And to be able to like go off go online and like maybe maybe like use the internet really well like almost like a robot would where you can like move the mouse and use the keyboard. That's what
[00:04:35] you have to do to solve like general purposeness for around a computer is you you can't rely on API or MCP. You have to figure out how to like navigate like a human can. Now at Hark, we've like we just released our first uh kind of model and research preview um last week. It's really hard for us to find now something that we tell it to go do on the internet. It can't do. >> What did you guys do differently than the other because everyone's trying to do computer use, right? So like I think Elon's got macro hard and chatd had their computer use thing. Everybody's doing it. You guys feel like you've
[00:05:06] cracked something. What did you guys do differently? >> Okay, there's a couple things we did a little differently. First is like everybody's tackling this from like using APIs and MCPs. Like the reason Open Claw got so great, it was like it could only it couldn't use the brows. It couldn't like go on and use Door Dash end to end because Door Dash has no consumer API. So we tried to figure out how to use like a how to look at a screen and one is we spin up a virtual computer for every agent. So they don't need like a MacBook or anything. So you can just spin up as many of these environments as you want in the sandboxes and uh and then you need to give it ability to like look at a screen
[00:05:36] and use like move the move the cursor and use the keyboard. >> Yeah, but I used chatp's computer use and it was doing that. I was like, "Hey, book a massage." And it opened up a browser and I saw the mouse going and it was trying to type the thing and it would scroll results. It was bad. It didn't work well, but it was it wasn't trying to use APIs or or uh MCP. It was trying to use the internet. >> Yeah. I don't know if I Yeah, it's got to work well. I mean, that's the whole point. But like if it goes in their funnels, it's like the whole point is like >> So that's what I'm saying. What did you guys do to make it work well? Was it like an algorithmic breakthrough? Is it >> It was in our post training like it was
[00:06:06] in our we have a reinforcement learning process that we think is maybe nobody else in the world has done. >> Well, get let's get some context behind this. Okay. So figure that is shockingly easy to understand humanoid robots and that business is going to be massive if it works. If you can crack the code, I think you said there's unbounded uh demand. Hark, I don't entirely understand what that is. Can you kind of explain like I'm an idiot? Because Sean, you should see I got the deck and it was just you talking for like an hour in
[00:06:37] front of a screen and then there was a list there was a list of a team and it was like a hundred guys who just moved here from China who had like the greatest backgrounds ever and you it seemed like you pretty much just raised money because the team was amazing and uh that that's all the deck was. It was just you talking in a video. Well, I mean that's kind of all we had of time we started. So, okay, what is >> I think the best way to become successful is to see how other people did it. Whether you're going to copy them or just use it as inspiration because then now you know what's
[00:07:07] possible. So starting at the age of 24, I did this relentlessly and I was very methodical about it and I created a spreadsheet where I tracked roughly 50 people who were uber successful and I looked at the year that they were born, the year that they started their apprenticeship, and then the year that they started, the first thing that made them successful, finally the year that they broke through. And I aggregated all this data along with the stories of what they did to be an apprentice and what they did to finally break through. And I put it together in a database. and HubSpot went and found this thing that I
[00:07:38] frankly even forgot about, but it did change my life and they resurfaced it. They made it even better and they put it into a thing that you can download for free right now. So, if you click the link in the description or click the QR code right here, you can see this database that I made when I was 24 and it changed my life. And so, if you're looking to become successful or you're already successful and just want some more inspiration, check it out. I strongly believe like AI will head in two directions like uh like and then at some point maybe even like maybe like head together like the first is we'll have AI out in the physical world that will like do everything in the in in the
[00:08:09] environment for you like laundry, dishes, cooking like run the supply chain and be in healthcare. The vessel for that is a humanoid robot. It's just a human form and it will just go out and do like you like want one piece of hardware that can like you know the hardware is capable of doing everything and you put like smart AI into it and it'll go off and do everything in the world. That's what Figure is working on. Separately than that, there's going to be this like really close like digital like AI to human symbiosis that forms. You're going to have like this very special thing that you can like talk to that's with you everywhere you go that
[00:08:39] will know all your stuff, have access to all your memories, have access to all your accounts and systems and be able to actually go do things for like a superhuman assistant. It'll be like um maybe the closest thing is like Jarvis from Iron Man and it will be able to do like it'll be like super human in almost every way. It'll know everything about your life. You'll be able to access it at any moment whenever you need it. It'll be in the background helping you out at all times. If you're on a like if you're on a flight with like a long layover or flight with like maybe say a short short layover and you miss it, it'll like already have backup plans already help you like figure that out.
[00:09:09] Like it'll just be something with you everywhere you go. We don't have that. We have like really good coding agents. We have really good chat bots, but we don't have like something that can go off and like be my Jarvis. In order to get there, we need to work on the like model side. It's got to be just better than text chat. It's got to be able to use computers, have like basically near perfect memory, be able to talk to you like just like a human would back and forth. And uh we have to have vision in the system. You have to be like look at the world and understand what you're seeing with it. And I think secondly, you need to have um you need to fix the
[00:09:39] the interface to AI. You have like um AI over here and a human and you have like an old hardware system in between like a call like a MacBook or iPhone. They were designed 20 years ago. They're complete rubbish for AI. They're not the right interface. So, we went out and we are out there designing what we think comes like after the iPhone for AI. And it's like an upgrade cycle. We see this all the time in startups. You guys see it, right? Like we're we're in an upgrade cycle with the computers and phones. They're just going to go away. They're going to be a new ones. They're
[00:10:09] going to be all AI computers and phones and systems and they're going to be great. They're going to be all real time. You can always access them and you want they'll always be like understanding what's happening. They'll always be able to reference things what's going on. You'll be able to abstract away most apps. You'll probably not have an app store. or you probably have an AI operating system. Uh it'll be perfect for you. You'll ultimately have your own weights on your own devices that you'll own and have with you everywhere you go. It'll be like a really great uh pairing. And we hired an incredible team. Teams like you know maybe like 80 or 90 now. Uh the guy that leads uh hardware design ABS previously
[00:10:41] designed for last several generations of iPhone, MacBook, MacBook Pro. Like he's just like the he's a stud. He's great. So, we're designing what we think are the next generation of AI devices that will kill the phone and computer. And then, uh, we're designing the next generation of AI models. The models need to get a lot more multi multimodal. They need to get a lot more expressive. Like the text encoding is just not enough for us to like really have a like a like a real AGI feeling with AI. So, we're working on that. We have our first AI pre we did our first research preview or computer using agent that we came out
[00:11:11] last last week. I think we were like top on some of like the leading like, you know, browser computer use benchmarks in the world. And uh it'll keep getting better. This will keep getting better and better. Like every month we'll just like it'll be better and smarter using a computer and faster. We're working on a couple other different types of technologies internally on the AI side. And then we'll launch the ability to use HARK on like traditional browser and iPhone and Android in about a month. So um you'll be able to start using it and then we'll have hardware coming. Um we're working on now. We actually have a hardware in the lab now we're using testing and it's it's crazy [ __ ] Like the stuff is like a sci-fi movie
[00:11:41] hardware. >> What do you think those devices look like? You know, people have been speculating because Johnny Ive, you know, got his his his shop got acquired by OpenAI and you've seen the videos of the puck and then the like this little puck and then there's like an earring. I don't know if that's real or if that's fake. There was like a leaked commercial for the Super Bowl again. Is that real or is that fake? What what's the story of that? And then what do you think these devices end up looking like? Are these watches, glasses, something else altogether? >> I think I've like really changed my mood on this a lot the last like year or so. But we have a really strong opinion here
[00:12:12] internally. Our opinion is that we what sits in the middle is devices that could possibly reach a billion units a year in the world. The only kind of things that we have like that in the world right now are computers and phones that kind of meet that like call like mega call like mega devices. And then you have things on the ancillary around it like orbiting this like big thing that are like Airbods, AirPods and you know like a watch or things like this that are like they don't sell a billion units a year. They're like 3% of like Apple's revenue and they're like they're they help the
[00:12:42] ecosystem as a platform. What we care about at HARK is trying to solve what's in the big middle piece. To solve that, you got to take down the computer and the phone. There's no way around that. So, you have to rebuild a new computer or new phone that's better and replaces your existing systems end to end. And then what's around there is things that like you will have we will even have at HARK that are like a helps with a family of devices that are um not a billion units a year but important for the ecosystem. My understanding right you're you're kind of saying the next the next
[00:13:13] device it might be like a phone is just going to be an AI native first phone, right? You're not going to try to change the form factor. >> No, I'm not saying that at all. You're going to want to like really radically rethink everything. Uh the like the first version hardware we have now in our lab is like unlike anything I've ever seen in my whole life. >> Okay. >> What lives outside of here on the edge are like like glasses and pendants and wearables and things. They're not they're not the main show. In fact, like the metagasses are probably one of the worst products I've ever bought. They're just horrible. They're horrible. Like I I can't even like figure out how to use
[00:13:43] it. It doesn't have its own network. It piggybacks on the iPhone network. It means your app needs to be open on your phone. The pairing's long. Like it doesn't work well. Like I can't think of any reason why I would need this thing strapped to my head for 14 hours a day. Like it's just like the wrong device. It's it's not like the end state is BCI in the brain and we're going to have like AI language devices for the next 10 years before that. And that like that's that's the path and it's not glasses. Glasses I think I don't even know if glasses will make our top like 10 list of devices. When you're when you and
[00:14:13] your team are like brainstorming, do you have a framework on how you can think outside of pre-existing norms? Because when you're talking about like I I literally can't imagine at all what you're talking about >> let's let's get down to like the substrate level here like first order what what has changed what's changed is we have like a new type of computer which is like I think of AI as a new type of computer new type of automation that's here that automation can do a few things that are like like when we're designing this we want to design around like key principles that could be like
[00:14:43] 10x better if it's like one or two times better your phone or computer you're not going to use it it's like literally 10x better what are things now that a deep learning brings that are like 10x better. There's a few of them like one is AI can like basically now like think and use computers and systems for you just like a human can. It can like talk to you. It can like see has like visual understanding. It has a real-time speech to speech. It can uh it can use computers and systems for you as f close to as fast or around as fast as a human can. Over time it'll be just as good as a human and faster um in terms of success rate. So you have a system
[00:15:14] that's like almost like humanlike in capabilities. It also can like have memory, meaning you can put memory into it. It like won't it won't it won't forget anything. I mean, near perfect over time. So, you have a system that's almost like a human in a box that has all the same like affordes a human has. And it's almost like the ability of like you almost like if you could bring a little human around with a computer on your on your shoulder everywhere you went, that'd be insane. Like would just like it was only for Sam though. Only Sam could see it. Only Sam could talk to you and only was like there to help with Sam. And that was like your whole life. and is going to be smarter and better
[00:15:44] along the way and had perfect memory and could use computers and talk to you and see you'd be like damn that thing would be like like it would be like be able to do anything you do on a computer. Okay, so your first step with your team is like just like let's just get rid of like any constraint ever. What would be the coolest magical thing if we had like a little guy on our shoulder that was AI all knowing and could see and hear everything we see and hear and then give advice to us >> like like what is the thing that's going to bring that's going to fundamentally reshape all this. >> Okay. And then from there like we got to like we got to design around that
[00:16:14] system. The competitive advantages here are that it uh is humanlike and capabilities and it has almost near perfect memory can go back and reference over time. My phone doesn't have that. Like I put a contact on my phone like last week and I was like I was like busy when I was like putting the phone number in and like a day later like somebody's like hey did you did you call that person? I'm like I don't even know the name. I forgot. I can't even ask my phone. Like it's just like it's so stupid. Like the whole system is and then I go in there like order Door Dash like a monkey. like every day now I'm pushing things like I don't do any of that now with Harkc it does it end to end for me on my drive to work I just
[00:16:44] like say order me coffee and it's just done it does it all for me in the background I don't have to touch anything it's all abstracted away and it's like if you had that little human with you everywhere you go you would just say like you would even predict probably Brett you want coffee today and be like ah yeah I do like let's get let's order but you know what make it a double shot today and you know like route it to the hark office instead of figure like I can like I would just and done I got it let me take care of it I was think like a monkey on my phone for the next like three minutes like trying to do checkout Door Dash. >> It It's almost like the phone is like a tool in the It's like a hammer, right?
[00:17:14] If you want the hammer to do anything functional, you have to pick up the hammer and start swinging it. Whereas the next generation is basically like having a handyman next to you at all times. And so you just tell them, "Hey, can you fix that window?" Just go fix the window. You don't have to pick up the hammer and start figing how to use it. >> Start start there. And then from there, you got to rapidly prototype. So when you come over, like we have like we've we've designed everything you could possibly think of. We 3D printed it. What were the designs that didn't work but were kind of cool? >> What were designs that didn't work that were kind of cool? The thing is we're
[00:17:44] building like many different devices now that cover like a pretty wide area of this. We have some pretty crazy stuff that we were like designing. So like it's not like you look at that you're like that looks like a that looks like this and it well that does over well over here. So it's like it's not as easy as drawing those parallels. It's like pretty quite radical. We we rapidly prototype all this. We have like a fabrication facility that does this stuff. We like we have a whole design studio where we work on this. I like you like use this stuff like over the coming like weeks and months. I'll like either carry it around with me, wear it, whatever. We end up doing it and we'll like kind of down selection. I we we had
[00:18:16] like one of the biggest telecom CEOs in in the world here that actually helped um with with the work with like Steve Jobs on iPhone 1 and he was here two weeks ago and he just come from meeting Tim Cook. Uh you know Tim Cook's on his way out as Apple but he's like was over there at Apple and came over here and we he saw our stuff and he's just like holy [ __ ] man. This is the first time I've ever seen anybody that could possibly take out like take out the big guys. >> Well, is it true to say that with like Archer, Figure, and Harkc, the the hard problem seems like can I just massproduce this?
[00:18:46] >> The hard problem is not that. We think we believe now the most important constraint to really solve is like building a really intelligent robot system to the world. Like there's a bunch of robots you can go buy now. You can buy some from China and you get them and they're complete crap. They can't do anything. They like you can joy sticking around. That's all you can do. And you like hit a button and it waves. It's got no hands. It's got nubs. And you're like, "What do I do with this thing? It's a toy." It's like It's like It's like early when I bought a I bought a DJI drone like years ago and I was like playing around with it. Then like a day later, I was like, "What do I do with this thing?" >> And uh it was like it was like hard to
[00:19:17] set up. It didn't really work well. Like you know, whatever. It's just like I flew into a bunch of trees. It just didn't work. I was like, "This is what am I doing with this thing?" Robots are like that now. Like where you can we can go manufacture a ton of them, but like if they're not really smart, like it's not really going to be that helpful. We're trying to crack like the true human level intelligence of figure. Like we really want to tackle like how do we make it so I can put it into any home. It can do every every every job I'd want it to do. That's what we're working on. We think that's the largest like like you know think about the largest like gap in the schedule of what we need to go solve for. Like it's that then beyond
[00:19:48] that like you know people generally sometimes confuse like consumer electronics manufacturing with car manufacturing. There's no company big company in the world that would look like say like uh uh I'm scared of manufacturing this consumer electronics at high rate if there's so much demand like this is just possible to go do I mean you can make them you make we make a billion phones almost like you know pseudo by hand in the world and with some automation but cars is a different story cars like you will die trying to manufacture cars there's like there's like lot of like you just like it's so and having seen like you know BMW is a
[00:20:19] commercial customer of us I haven't been in BMW and a few other groups like it's gnarly early. The reason why cars are so hard is that you can't hold the part in your hand. Phones you can just like always hold in your hand and go change or whatever, move and hold. Like cars you can't you physically can't. So you need robots that like literally pass it to other robots that put things on the chassis. And if any of those break across like thousands or 800 robots, you're dead. It is you're the whole line's done. And so it's just like this like huge giant robot you're building that's building the car. And with figure you can hold any part in your hand. So I
[00:20:50] think we're like if we're like between cars and like consumatics, we're like over here closer to like you know we're like you know the 40% level over here by like cell phones like we you know we just made our 1000s uh EVT robot for figure 3 last week or week before that. >> When you say you made a thousand those are a thousand that go to customers like BMW or you're making prototypes internally. What does that mean? >> We have like two bit like large customers. Uh we have us as a as like an engineering and like AI research org that needs like robots like here like
[00:21:20] every engineer needs a robot. We need like every lab needs robots like we need like do tons of testing. Uh there's just a lot of work we need to go do uh internally. We call like maybe like engineering fleet we need to go to and the second one is go to customers. So we have going to both right now. Uh we've actually shipped out robots to our third customer this this this week. >> When they go to customers what do they do? What what what can the robot do? what you know maybe can't it do at at this point. >> We do a lot of like logistics stuff right now and packages. Uh we have other stuff we've done in manufacturing mostly just manufacturing logistics just we've
[00:21:51] done in the past. Um but like we're also talking to folks about other industries. >> And at this point when it goes to a customer and it's doing I don't know what you said like packaging work or what is that like sorting or tearing or what is it doing? >> They just did a live YouTube video and they had hundreds of thousands maybe millions of views of people watching this robot sort packages off of a conveyor belt. >> Yeah, I saw that. So is is that the type is that like would that give me an example of one of the jobs just like >> that's an example of like like a very close like one of the one of the works we do
[00:22:21] >> is that customer like oh this is awesome because I can't find the labor to do this uh it's too expensive to pay humans this is way cheaper or is it just like hey look today it's not faster cheaper or better necessarily but like it's an investment in the future where two years from now that cost curve is going to work and it will be faster cheaper you know whatever. No, no, no. It's like uh it's the pitch is like they come to us and they're saying like we're dying with labor. It's like we're like we have like really high turnover. Some areas have over 100% turnover per year. It's really expensive to find talent. We have like a
[00:22:51] just a large talent shortfall. The talent's really expensive. Like wages are going up and we like we don't have a solve for this. We can't figure out how to automate all this work and uh we need you to come in and help us. We have an ability to make a lot of good money in our contracts and the customers make like really good ROI on this. Like you got to think like a robot can do like multiple shifts per day, work seven days a week. Like we can like have a lot of uptime. Uh the task you saw like on the on the on the like um logistics line that we should live stream was actually a real use case for one of our
[00:23:21] customers. Uh that needs to be done at 3 seconds a package and it initially needs to be done five hours a day. Like I think it's like 5 days a week. We did that 200 hours straight at 2.9 seconds a package. So, we're already at human speeds. We're already doing this here now. They're already having ROI and uh we're now in the early stages of like getting these out to these customers and scaling it up. Uh over time, it will just put billions out to these groups. >> Can you help me with like the kind of truth first fiction? Cuz uh one of the weird things is as an enthusiast or a
[00:23:51] lay person who's who's excited about this future, you can't really it's like really expensive or hard to test this, right? Right. So, I'll see like a a Chinese robot and it's 20 grand if I want to buy this robot. I have no idea really what it can do. I see uh you know Elon will go out there and say, "We're going to we're going to build a million of these things in the next year. We're going to ship them." Uh then you get like 1X and they're showing their hand and they're like, "Look at our hand. Look, this is a this is the best hand you've ever seen." And then there's this service in San Francisco where they'll send a robot in to clean your apartment and they're like, "Yeah, that works today." So, can you help me separate
[00:24:21] fact from fiction? and it seems really hard compared to most categories where I can just try the products quickly online or buy them and and and test them out. >> One is um the amount of like noise in the market for a signal is just like it's like like you mentioned is like it's like it's it's out of control. Like the there's just so much [ __ ] out there in the market. It's like really hard to tell what the hell's going on. So, let me summarize what I think is like the most important and work backwards. >> Okay. What I think the most important thing to do is uh to be able to ship
[00:24:51] robots autonomously at scale in useful work environments like they can like you know cook your dinner like like clean your dishes like um make your bed like run the supply chain end to end work in healthcare build a building like do logistics like that sort of stuff that stuff requires uh fundamentally onboard AI you can run so you can do like autonomous work you can't solve it with code you need to do it autonomously. You need to do over long periods of time and you probably need to move around and use
[00:25:21] like something in your hands and move move stuff through the world. You know what I mean? It's like really got to like do do stuff economically like move got to move like electrons around. So I think at a high level like what we care about is not like the best robot that's doing back flips and running the fastest mile or dancing or in a parade or running outside in the woods. Like you know we don't care about that stuff. Dude, I can't wait till I see a figure like on a smoke break at the BMW fact. Like, [laughter] I could have been a great back in high school, but I blew it. Now I'm working at BMW factory.
[00:25:52] >> I've made jokes with you before where I was like, you started with Veter, which was just like a job recruitment thing. Now you're on these world changing things, and you were like, well, Veter actually is world changing, and here's why. And you gave this pitch. It was very good. You're very good at pitching. You're very good at raising money. You're very good at um being charismatic and convincing people of stuff. When you're crafting a pitch to recruit and convince people to ch change their lives, to uproot their lives, and to trust in you and to come and build a company, how do you craft that pitch? And what was that pitch for some of your companies?
[00:26:22] >> I mean, most of all, these are online. I mean, the figure master plan is on the on the internet on the site. Archers was up for a long time. I posted about it. Like, I think like deep down I really want to find folks that really care and are obsessed. And I'm like, I think most of my time is not I know you want to know about the pitch. Most of my time is trying to find those folks. I found that even in the Bay Area where it's probably like the richest AI and engineering like uh folks in the world, 90% of everybody out here is not good at their jobs.
[00:26:52] >> How do you tell who's good and who's not? >> I technically assess them. >> All of them. >> Yeah. >> To do that, does that mean you need to be as good or better than them technically to be able to assess somebody? >> I need to know like a certain guiding principles. Like for instance, I need to know like if a if you did the work or if you like watch somebody do the work. If you've done the work, it's like it's like it's like a scar you carry with you. It's like it's like dug into you. Like you know all the details. You can talk about it freely. You don't need to think. You'll understand how to like reverse engineer everything you've done
[00:27:22] and discuss it. The folks that haven't done it can't do that. They just like they can't even go like they get one layer and they just like instantly blow up. They can't talk about it. They don't know why. Out of a hundred candidates who sound good, how many like their resume looks good, the recruiter thinks they're good. Out of a 100 candidates, how many do would you say actually hit that bar? >> I'll give you an example. We have like a really challenging process to go through to be a mechanical engineer here at a figure. You have to be able to build like actuators from scratch. There's bearings and motors and, you know, we
[00:27:53] have a we have a we have a gearbox. We have like other sensors inside the system. It's a really it's very compact, you know. Uh it's just a very difficult thing to do. Uh and like really hard requirements. We've been doing 10 case studies a week for six months and have not hired anybody. >> That's insane. >> It's insane. >> But when you do get someone qualified and their competing offers are companies that are larger or or more liquid than you and the offers are I I I I think
[00:28:24] they're like tens of millions of dollars a year, right? The AI side is certainly like that. Like the AI side has gotten um and it's it's mostly all driven for meta. Like at Hark like I've never seen I thought maybe like Meta was like paying these people for like like a year ago and it was like would go away. They've not stopped. >> So what are they like what's a crazy story that you've heard? >> I think we gave an offer to somebody that was really senior that was like they were coming from X8i like XDI completely blew up like everybody just left and about it 6 months ago. It was just like every it was just like like
[00:28:54] macro hard like got fully disbanded like there was basically a bunch of stuff that happened. We interviewed a pretty senior guy on the AI infra side. It was great. I think I gave him like a really good package like of series A stock at HARK >> and it was I don't know I don't know 1520$20 million of stock >> over four years >> over we do five like for uh my companies in the early days and we transition to four a little bit later. We're still a five. And uh you know I was like I think we're like a I think we can like 10x Hark here pretty quick. And so I was like okay you have like you know 15 20
[00:29:24] million. I think 10x you have a few hundred million. We 10x one more time you have a few billion dollars and I think we can do it. I think we like we have to like obviously it's going to be hard but I think we can do it. And he got an offer for to go to Meta for 36 million of four years of our shoes. And he's just like it's kind of guaranteed cash. you know, I go there and I have to like weigh this like maybe like $200 million at HARK or $20 million or maybe like 36 for sure at Meta and he left and going to Meta and they've been doing that like every candidate we sp speak to is like making some absurd absurd thing.
[00:29:56] They just haven't stopped. They've just been they've been at it since like for like a year a year and a buying talent. They've been buying their way into the AI race. >> What What do you think of that strategy? Like you know even if you kind of hate it, do you respect it? Do you just think it's a fool's higher end? What do you think of that? >> I really like it. I think like the AI space is what I found is the folks that really understand how to do like language pre-training and mid-training and post- training especially pre-training and the infra around supercomputing and data and evals and all the right stuff you need to get put in place to do that right and the amount of folks that really understand the
[00:30:27] right kind of like recipes that transformers do well in in you know arounde or whatever you're going to look at I think is really hard to find it's actually really hard to find the actual folks that know what they're doing. I think there's probably my rough calculus now is probably like a rough back of the envelope is probably like 20 to 30 people in California know how to build really good AI models. >> Wait, so but is that trickling down? So you said that there was a senior guy, but like are even some of the less than senior, the 20somes, the young 30omes, are they still getting eight figures a year? >> No, the like the junior guys like the
[00:30:58] guys in like their 20s and like the like late 20s or something are getting like they're making like a few million total. So they're making like 200 250 in base. They're making like another million or whatever like in in a year in like our shoes every year. And so they're going to pay like a million to like you know like 750 to like 2 million or so range per year. >> And that's uh that's been driven up by Meta and but then all the other labs have have like have like followed comp. >> When I asked you what do you think of
[00:31:28] that you said I like it. Were you being sarcastic or you're you're saying no actually that is smart given how hard it is to get this talent. I think it was really smart and I would have done the same thing if I was I was I was Mark. I would have bought my way into the race and I think he's like he's doing that now. I I don't think I would have done that. I want to understand it and I want to like first order like find the right folks that really care deeply about this and not hire like like mercenaries >> and so he [clears throat] hired a bunch of mercenaries. They're just purely money driven. He they came over there. No other nobody wants to go to Meta.
[00:31:58] They just they're going there because they're getting paid a guaranteed RSU package by sitting around. And what's happening is like you don't need like a thousand people or 500 or 300 to design AI models. You need like a really good team of 20 or 30 or 40 people. And that you can get there without doing this. And those people probably would care more deeply about the mission and where you're at and be more committed than just if you purely throw money at the problem. But I think if I was like I think it was a really good strategy and it's working. I think hats off. like
[00:32:28] really good execution, their recruiting efforts and how they're structuring this stuff and uh it's like I think it's I think it's like paying off for them. Juryy's still out if they can like actually ship real products. I think like the problem I have with those groups is they've just traditionally have not been able to do things new. Well, I mean I think Facebook is probably Meta is going to go down like one of the greatest acquirers in all time with like you know Instagram and WhatsApp and different way they've like bought their way into those spaces. Uh but like you know if you look at like the Ray-B bands and everything they're
[00:32:58] doing is just like it's it's not great work and so I think the question really is how do you really do great work here? I think like we're even talking like we're using like the heart system right now and it's so good. It's so much better than anything I use today. >> You got to send it to us. >> Yeah. Can we use it? >> Well, yeah. We get you guys early on. Yeah, for sure. >> It's like research preview. There's like 500 PhDs and then me and Sam. >> Yeah, exactly. No, we like like every other platform is >> Park. What's the weather outside? >> I can answer that. Yeah, no problem. Like I I think what I'm trying to say is like every week there's like five or 10
[00:33:28] like junk AI slop startups or like things that are coming out. They're just like not very good. Like this whole space has gotten to a point where like there's just not great things coming out the door. I think the stuff in coding is probably really excellent right now, but everything beyond that is like just kind of like not great. On January 1st of this year, you've made four predictions for the year. I want to check in and see how how you think they're going. First one, uh, number one, humanoid robots will perform unsupervised multi-day tasks in homes they've never seen before, driven entirely by neural networks. Long time horizons going
[00:33:59] straight from pixels to torques. How are we doing on that one? On track, off track, or done? >> On track. >> On track. You got four months. >> Yeah. I see every day like what we're doing. Like we're on track. The hard part here is um we already do pixels to torques. [laughter] It just means like we're taking camera feeds and we output like where to put the motor like put the like you know we want to put a we want to like tell the motor like what to what to do to get to the hand in the right spot or the joints. So we're ready to do that. Um getting into a new house never seeing to
[00:34:29] do work. That's the hard part of this problem. Um we're working on that. I'm working on that every day. This where I spend about 3 four hours a day every single day seven days a week on this problem. >> So if a figure robot showed up in my house what would it what's the bottleneck right now? Like it wouldn't know what to do. It wouldn't know where to go. it wouldn't be able to, you know, fine-tune, handle my dishes. Where would it suck for me? >> We can fold laundry, like as an example, but then going to a new place where we're folding in different location with different lighting and maybe different like table height and different types of laundry and different like types of scenarios it's never seen before. It's
[00:35:00] like the model is like out of distribution. It doesn't know what to do. It's like if you removed all the pyramid data from the pre-training of LLM, it wouldn't know how to talk about pyramids. >> And we just like we don't have enough of that data out there. It's not on the internet. So you have to go out and collect it. So what we need to know is like how much of that data we have to go sample in the world to be able to train the model to be able to go into your house and say fold clothes is a good example. >> Hey, stupid question. Why do all the robot companies care about folding clothes and and doing laundry?
[00:35:30] >> Wouldn't it be commercial better just to say, "Hey, we're going to build like the best warehouse worker cuz there's already 20 million of those in the world and that represents this much billies." And of course that buys us the runway to like get the robot folding, you know, robot done. But like why do you care about that at all today? Why not just industrial work that people don't want to do, companies need done, they're ready to pay, and it's not like my home where there's all these other sensitivities. Why do you guys care about that right now? >> Uh we didn't care about it in the past. We like when we first launched, we're like, we're going to basically do the
[00:36:00] commercial side to pay for the home long term. And uh that was the strategy. It made a lot of sense. Like there's like we can charge a lot more in the commercial market. It's like much easier to do. It's like lower varitability. We're in like a little work site and just work 24/7. Just so much simpler. >> Uh what I've learned now is that the home is super solvable today. >> So like we can like not go work on that problem and just like sit here and work in a warehouse, but me or none of my guys want to solve that problem. We want to solve a robot that can go into any environment just through language and do work. We want to be the first to do
[00:36:31] that. You can probably do that with a 100 robots and a 50 person team. So that that company overnight would be a trillion dollar market cap. >> That sounds good. Do that. >> We're doing [laughter] that. That's what we're doing. Like we're going to solve that. I think we'll be the first. We call it like solving general robotics. >> And I Robot. Don't they attack the humans? I don't remember this movie very well. Just >> Yeah. Don't worry about that. >> Okay. Not that part of who could win. Who could win in a Who can win in a fight right now? Can a Can a human still win? >> Yeah, human can still win. >> Okay. Uh what are the other predictions? >> Other prediction um one you had on here.
[00:37:02] Daily AI usage will shift. People will move beyond text to highly multimodal. Voice agents with permiss uh persistent memory will become common which will push AI closer to the synthetic human intelligence we've imagined in sci-fi. >> We're doing that at Harkc. We'll ship that in a month and our first version of it. It'll get better and better. Uh I think we're on track for that. >> Have the labs ever like has Chad GPD or Claude, have they ever released the data on this? Like I use a ton of the voice thing. Sam, do you use the voice stuff a lot? >> Yeah, I don't type really at all. >> Yeah. I wonder it's probably already a huge percentage of gotten to the point
[00:37:33] where like offices need to change. Like these open air offices that are like popular in startups, they're kind of whack right now because like I want to talk in private. >> Yeah. >> Yeah. A lot of engineers have microphones now where they're whispering and [laughter] they're just like in hushed tones whispering to their computers. >> Yeah. Like I didn't I was like talking last night and I was like, "Claude, why am I so indecisive?" And then my wife was like like she was like like damn, dude. She can hear everything I'm talking to Claude about now. Yeah. No, I I I I talk all the time, but it's embarrassing.
[00:38:04] >> Yeah. Like even speech still like sucks. It's still not great. Like it's um this like almost like you set up you have to go there, you have to turn on like it doesn't like really remember what you just talked to it about. Like it can't do tool calling and computer use very well. Like it's just like limited in these like you have to use it for a certain session. I don't know if we'll hit it this year, but certainly in 2027 you will hit like a full human touring test with speech. You'll be able to take a phone call from an AI system on your phone and I'll be able to fool you guys. I'll be able to have like a human call you and a robot call you and I don't think you guys will be able to tell the
[00:38:34] difference. Uh that's that's a 2027 event I feel pretty strong about. >> All right. What's the third and fourth? >> You had over the past 10 years, school shootings have increased by 10x. Uh in 2026, the first full scanning system capable of detecting weapons from a 20 foot standoff will be built and beta tested in a K12 school. Ah man, we're going to be we'll we'll have our first we're building our full scale system starting in October and I think it'll I think we'll bring it up before end of year. I don't know if it'll be at a K- through2 school. So we might miss this one by a quarter.
[00:39:04] >> Do you have separate CEO running that one or you're the CEO also of that company? >> I have a chief engineer from JPL and NASA that's like really good and it's mostly a pure engineering project project. Uh there's like really not much to do on the business side. Like there's, you know, we have like some supply chain stuff and other things, but most of it's just like purely can you build a system that can detect weapons. Well, it's partly like a hardware problem. It's partly an AI problem. It's like roughly like a large scale. It's like in like a deep deep tech, deep engineering problem to solve. And my whole team is just all of engineer.
[00:39:34] They're really good. We actually made a pretty big change of cover. We would already be in market by now. And I like I pivoted the whole technology system about a year ago. We were building this like we basically I I found a way to do everything very cheaply in in silicon and chips and reduced the price by like 90% make it much more scalable, make it work better and we pivoted. The problem was that the fabrication times for designing our own chips and getting them out took about a year. So we just got those chips in like
[00:40:04] a couple months ago and we're testing them and they're awesome. Uh now we need to make more and there's another six month lead time to make even more of them. So, it's just like we're like dealing with like real silicon long fabrication of very difficult chips lead times now. Um, we'll be out of this at some point, but uh it's not like chips you can go off and buy at, you know, off a shelf. These are like custom design cover chips that like nobody's really ever designed before. We had a special fabricator in Europe that had to go make them and uh it took about a year. >> Hey, you are firing on all cylinders
[00:40:34] right now professionally it seems. And I'm and I'm and I I actually would like to know like what's the trade-off for the life that you're living right now because you're very optimistic. You seem excited, but what are all the trade-offs? >> Yeah. About 5 years ago, I had like a like uh you know like having kids and the companies. I had an issue where like I I think of my life is like three pockets. I have like work which I care deeply about. My family I have like three kids. They're pretty young right now. And then I have like they call like the other stuff where it's like a
[00:41:04] friend's in town or you need to go to the annual golf trip or like it's a bachelor party or like you know it's a wedding in like Europe or whatever it is like in this bucket over here. And I felt like I needed to make a decision on like I I need to do like if I want to do any of these well I kind of like I can't do all three. And what I wanted to do really well is like family and I wanted to do like business stuff. I wanted like I want to be like A+ in those areas. And so I basically stopped the third bucket. I don't like I don't like do anything anymore over here. So like a I had a
[00:41:35] friend in town from a college was like my like freshman roommate and he was like I'm in town for 10 days. I'm in the Bay Area. I want to meet up. I haven't seen him like you know for a long time. He'd be great to get a coffee. I was just like man I'm going to be real. I don't have any time. I can't I can't meet you. He's like I'll make myself available. Come to you. I was like I literally have no time. Every minute I'm away from one of these two is a minute I'm away from my family or work. And there's almost a limited amount of time I can put in both those buckets. >> Can I ask you about your workflow? You made a joke. You're like, I don't use Slack. If you're comfortable, could you just like hold up your phone right now?
[00:42:05] What's on your what's on the home screen of your phone? What what what's your app setup? What do you got? >> All notifications. >> Well, you got to open it up. >> Oh, what's my >> So, you have just tons of texts. >> This is like my uh those are all I think Slacks and texts. Uh I mean, I use Slack. I just I can't get through it during the day. I have heart going through it and then they Hart texts me, "Hey, it's important. and I need to look at with a link. >> So, what's your setup like? What's your like dayto-day when you're do you use a laptop at all or you only on the phone? >> I use a laptop. Yes, laptop a lot. Uh
[00:42:35] laptop and phone. I would say I use Hark now for all my AI stuff end to end. Um even like tracking like stuff I'm doing on engineering projects, recruiting, all of it. I do it track my it's in my email, it's in my Slack. >> What about your to-do list? >> That's all in HARK. Hark managed all that. >> So, what about before HARK? Um, I would my to-do list was done in a Google doc. I had like a docker called replanning and it would constantly keep updating every week. I would come in on Sundays usually and update my plans for the week and I updated there.
[00:43:05] >> And what about health? Are you doing anything for health? >> Yeah, I do like um I've done I've gotten like access to like some special doctors and things now where they basically send you through like the quarterly blood tests and like the the whole body scans and like the CT scans of the heart, everything. And it's been honestly unbel pretty pretty unbelievable. >> What was unbelievable about it? >> The amount of data you get back and the amount of thoroughess of all this like uh like for instance like like you know if you get a you can get a CT scan of your heart for like 100 bucks uh I think you can basically prevent heart attacks.
[00:43:36] You can get a full body MRI and I think you can like have early cancer detection. A lot of blood work and find some anomalies that you can go fix and better for your health. Um so there's like maybe like a dozen of those. >> Yeah. But the solution to all those things are probably things you're unwilling to do. It's like you're probably willing to eat whole foods, but like it's like get up, go for walks, exercise, and that was outside of your buckets of of of focus. >> Yeah. Unfortunately, I haven't been able to have enough time to exercise enough. But, you know, eat right. Like, I've I've like uh I eat pretty well now.
[00:44:07] >> Yeah. >> I mean, listen, like something's got to give. I can't sit here all day eating and like I got to go work. I I love I I like, you know, I want to go crush these businesses. when you uh you wrote in in like our prep doc, you said, "I went all in on my first three startups and I pretty much hit rock bottom every year." Can you describe what you mean by rock bottom and what is your method of dealing with rock bottom? What's the conversation you have with yourself or kind of the the entrepreneurial strategy you have when you kind of hit those
[00:44:37] lows? >> Yeah, I um I basically almost for like 15 years was like always running out of money. you know, at Veterary, we we had a couple pivots early on. We ended up raising like a $500,000 convertible note in 2015. I at that point, I think I took out like a 50 or $100,000 loan. I was not paying myself a salary. I was in New York City. I was like so broke. I was in the negative. We raised a convertible note. It did not look great. And I think it was like 6
[00:45:10] months later we launched the marketplace at Veter and it just like completely took off. And then a year later we sold for 110 million. And I think that period from 2012 to 2017 was just like was like I like I had like basically like debt. Things weren't working and it was hard. >> And what's what's the inner monologue? What do you tell yourself? >> Uh the inner monologue is like this really sucks. Super painful. I think at that point you just got to go like day for day. You just got to make it like day. You got to make when things get really bad like that, you got to build a punch list and you just got to get
[00:45:40] through it. Like there's only way out is through. So you need to build a punch list and you need to get to dayto day. You got to go dayto day. You can't go week to week two days look at Friday. You look you got to go every you got to get to the next day. Pile through it. I was training for this ultramarathon and I hate like really long distance running and I read this story about this guy who kind of helped me and he was like just all you got to do is like pick something like it doesn't matter if it's 100 feet or half a mile in the distance even though you have 49 miles left to go in the race just pick something half a mile away and tell yourself once you get
[00:46:11] there then you'll consider quitting and then you get there and you're like okay maybe I have a little bit more and you pick another thing just like only 200 yards away you're like okay I'll consider quitting when I get to that. No, I was like great. I think it's exactly how I thought about it. But it was like it's like okay. So then I sold Very and then I was like doing Archer. I was like, "Oh man, I just made 110 million. We like 12xed all the adventure guys." And then I was like, "We're going to raise money. It'll be I'll be raised money. It'll be fine." And like everybody's like, "What are we what are you doing >> for Archer?" >> Yeah. Everybody's like, "What are you doing? We're not going to fund this. What are you talking about?" >> How do you fight that inner monologue
[00:46:42] where everyone says you're stupid and wrong and this is silly. Just go do software. >> It's coming from a place of conviction. Like I I know I'm right because I've done I've done the work. I understand it. I'm on the floor. >> Yeah. But the odds are still against you, right? >> But that's that's that's that that's the game. That's when you play this game. It's like you like sign up and like 95% of everybody around you will fail. Like I remember what Veterary We like we we started at the NYU incubator. I was like so excited. We got in like the one of the like the semesters and there was like I think it was like 50 companies that were there. We started in Soho. Um
[00:47:12] it was great. We had a great time. the I think if you look back like I think like five years later, me and one other guy are the only two people that made greater than zero dollars. One other one team 48 companies went to zero and I was just like holy [ __ ] If you're around this game for long enough, like everybody dies and that's everywhere. It's been like that since for 20 years now. I've been watching around you like you see all the tech crunch stuff and things on X about people raising money and all this and just like over time that all just kind of fades away and
[00:47:43] it's just really brutal. Um so I had to like I had to like I you know I had to like I bought a house and like I had to put all the rest of the money into Archer and then I had like a stock lock up. So even while I was coming over to figure like the stock was like unlocking I was funding figure with stock from Archer because I had no other cash. stock was coming down. The stock was just like literally like a falling knife. Well, at that point [laughter] it was just like I think it went from like 10 bucks to like two and since like gone
[00:48:13] up a lot but like uh I had to take a second mortgage out of my house to even fun figure. >> We asked you one of your philosophies and you said um I believe that doing hard things is easier in many ways than doing easier things. Could you explain >> like everybody's trying to do easy things. When you work on harder things, you have like less generally like overall probably there's like first order like less competition. You have probably like a hard thing probably means like it could be a potential like really big TAM, really big exit if it
[00:48:43] works. You have this like you know riskreward trade. You have folks that probably want to work on hard things probably like the best overachievers in the world that kind of want wants to work there. Generally like you know hard things have this like binary payoff for investments. they like really want to fund those things because we could have like a 100x return for the portfolio and I think there's like a nonlinear curve to scaling like here the difficulty here meaning like I think a lot of the hard things are not like 10 or 100 times harder I think the hard things sometimes are like two or three or four times harder maybe five times harder but they're not 100 times harder so you
[00:49:14] might have a hundred times better payoff but it might be like three or four times harder I'll give you an example in robotics I think like largely building like quadriped robots like four-legged dog robots versus humanoids. Like probably humanoids are probably like three times harder than that. Maybe four. That's it. But like there's really no I don't think there's like really a real market for for humanoid like like those dogs, right? I think it's just like a niche thing. I don't think there's a real business for it. And I don't know anybody really at this point in like really wants to spend a lot of time on that. So like you do humanoids, it's like okay, three times harder, but
[00:49:44] it's probably like a million times higher payoff. Probably like a million x or a billionx higher ROI for that. You know what I mean? for investors, for humans that want to work there and get stock and participate in upside and for everything else. Like what why would you ever like want to work on like like four-legged dogs? Well, what economic value can a robot dog bring at scale? Like if you really understand it, like I think there's everybody's trying to do the easy work and it just becomes really difficult, you know, look at look at all the AI slop like open claw harnesses out
[00:50:14] there today. It's all crap. It's all not good. They're all going to go I don't think any of them will make it long term. You might have some consolidation here and there for aqua hires and stuff, but like that's going to go all the way. >> Dude, you talk in so many absolutes. Has that not gotten you in trouble ever? >> I don't know. I mean, mark my words. Like like you think like like have you guys even used Open Claw since then? >> No, I don't know how to. [laughter] >> Do you use Open Claw? >> I I never I never trusted Open Claw to set it up. I was >> Yeah, I used to use it. I don't use anymore. It's not very good. Like just
[00:50:46] the wave's over. I don't know. I'm just I'm just trying to say like I think it's like I think the most important thing you can do as a founder is to think through what you're going to actually go do because you're going to spend the next 1015 years doing it and it'll it'll like it'll map the whole course the probability course it's like a probability weighted decision of like or probability of like of like potential outcomes. >> Well, but you're you're you're you're talking about a very particular game like like for example uh you're as you said you're like we're going to be a trillion dollar company or we're going to go bankrupt. Like it's it's binary. Most business is not binary. You're, you
[00:51:18] know, you're playing the game where binary is the outcome and that's what you like, but it's not like that for a lot of people. Um, like for a lot of people, if they can build a really cool $10 million a year business, that's a massive home run. >> Is it would like if like if they can do that well and you look back when you're 70 or 80, would they would you have asked the same person, hey, you built a really cool$5 or $10 million business, you did it for 30 years, you didn't do anything else, you didn't try anything else while you were doing it. You just work on that business. Would you have gone back 30 years ago and tried to take a bigger swing? >> Like would you have taken a different
[00:51:48] swing than Veterary? Veterary was like that. >> Veterary is my bridge. I sat inside of Veterary for like seven years. Like we literally built like a marketing automation tool for us internally. And then like a year later I was like, "Oh man, look at this. It's outreach.io and it was like a billion dollar company. We built that internally a year or two prior." And then like watching all this different stuff happen and I was like, "Man, we like we actually did some of this work internally. it's like value less than some other groups out there. Like this whole decision of like what you spend time on is like super critical. Uh for startups, assuming like
[00:52:19] there's a and I do think startups are like I think it is kind of binary. Even guys that get to $10 million, there's probably like another 90% of those folks that just didn't make it when they're out there trying. So I think it's just I think it's just hard. And I think dude, kudos to guys getting like five or 10 million in business. Those that's that's hard. Um especially doing that if maybe a little bit of capital or no capital coming in. Hey, let me ask you real quick about your your uh other stuff you've seen. So, I'm sure because you're doing really interesting work, you meet other founders that are doing
[00:52:49] interesting things in at, you know, unrelated spaces. So, non humanoid robots, but equally cool, interesting peak at the future. I think you've probably seen more of the future than than us and definitely more more than most of the listeners. Can you give us uh anything that you've seen or heard or read about a founder you've met that's doing something that's like oh yeah you you guys you guys realize right the future is actually going to look like this and we're just you know not it's not evenly distributed for all the rest of us yet. I like I like looking at I like trying to think through this problem of what was the world going to look like in 30 years or is there where
[00:53:20] everything's headed. I think we have an energy problem like not an energy consumption like like a generation problem. So, how we maybe both, but like ultimately how do we like generate more energy uh as like a species? I think there's like there's like a there's like a secular trend here that you want to go ride and really help. And I think there's um a lot of work done correlating this to like like standards of living for humans. So, I think there's a lot of work here on like what is the next generation? Is it solar? Is it wind? Is it nuclear? Like and then
[00:53:51] there's a bunch of different traits inside of here for fusion and vision and the rest. Like I think it's a really exciting area. I think it would take a long time, but you need like I think you need like really great entrepreneurs like there solving that stuff. I think AI is just going to dominate a lot of stuff in the next 10 or 20 years for all of us here. I think it's going to be like 100 we all live through the internet. Like I think it's going to be 100 times bigger than the internet. I think it's going to be so so big. It's going to it's going to AI is going to eat the whole internet. It's going to eat it all up. And uh I think it's going to be extremely like large trend both physically and digitally.
[00:54:22] What are there any products that you're looking at or companies that you're looking at now that are not already the mainstream that you think are good examples of what you're talking about? >> I mean, we're working on this stuff at Harig. Um, it's unclear. We're still in this spot where like it's really not clear who's going to do well here in this stuff. We're in this like foggy area for a lot of these stuff of like there's no been no breakout here. Uh there's been early wins and early breakouts, but there's like a next leg here that we're going to go through and we're like we're in it now. I think we'll know more in the next year or two
[00:54:52] what that really looks like, but um I mean I've used like every AI device out there. Haven't been super thrilled. Um I don't know if you guys are seeing stuff in the market for these type of things, but like I haven't like, you know, been like, man, this is like a crazy great product. >> I like the small stuff like I like Whisper Whisper Flow has like pretty meaningfully changed how I communicate. >> Yeah. >> Um that's been pretty cool. I think that's been my big standout the last six months. What about last question? What about um people who inspire you?
[00:55:23] >> I think I really admire the folks that are like fully dedicated in their craft. You like watch the Michael Jordan documentary. He's just like he's just like I just want to be like the best in the world at this. I think for like for startups have that same thing too. And I think first and foremost like you know what I can read I never met Steve Jobs but like my lord like stories I've heard and everything else the guy was just like an unbelievable operator and product led founder. Um I've also got to know Jeff Basos pretty well. He invested in Figure and he's been here a lot of times and I think uh I think Jeff's has has been a really good soundboard for a
[00:55:54] lot of things we've gone through. Um I had Jensen in here last week again like we are fairly close and I think Jensen's just an unbelievable operator as well. He's very hands-on has a very unique way of managing Nvidia and his organization last 30 years and it's uh I I think he's like I think he's done a lot of really good things. >> What advice did Jeff uh give you that was meaningful? Jeff said when last time he was here, he's like, "Listen, you're at a really interesting period because like you figured out how to do this somehow." In the next year or two, you're either going to figure out how to break through and really get this like
[00:56:24] working in a bigger way or you won't. And like this is like you're it's game time for you now. And you got to just get wired in and like figure out how to break out and make this thing work and scale it. And you're at a really interesting point. I don't know how you got here and I don't know why you got here, but you're here and you need to figure out how to like your next, you know, you're on the big field now and your next big push is going to like make or break it. So, which I think he's largely right. Like I think we're like we got robots now doing this stuff autonomously with AI throttles, which is crazy. I think four years ago you've been like I've been like no way. Like no way you could >> dude. Four years ago I came to your
[00:56:54] office and you just had a knee working and I was like oh that's a knee that's cool. All it was was a knee. You had like there was five engineers. You're like this guy just got done building the Tesla X or Cybertruck or something. This guy did this amazing thing. This guy cured cancer. Look how the knee moves and the ankle has dorsal flexion. And we were just sitting around looking at this knee. And that was like the coolest thing. >> I know, man. It's like And now we have like AI that's working on a humanoid robot. We're taking in cameras. It's doing inference on board. It's outputting all the joints go, you know,
[00:57:25] it's it's unbel it's unbelievable. And uh it's crazy. It works. And um and yeah, the next leg up is just like making that work at higher scale. So I don't think it's been great. I think there's um I don't know. I think those are kind of some like I think really good folks to look up to that really like like love their craft deeply and really care. >> Well, Brett, I think it's time for you to get back to work, my friend. >> Great. Thanks, guys. It was good to see you again. >> Thank you so much, dude. All right, that's it. That's a pop.
