Intro
Jason Calacanis and his editorial director Lon Harris pull apart Nvidia's agreement to buy Hugging Face, the open letter more than 100 companies signed about AI-enabled cyberattacks, and the Wall Street Journal op-ed published under Stanley Druckenmiller's byline. Halter's Craig Piggott and Hop Aero's Matija Milenovic take the two founder segments.
Host: Jason Calacanis, founder of LAUNCH and host of This Week in Startups
Guests: Craig Piggott, founder and CEO of Halter; Matija Milenovic, founder and CEO of Hop Aero
Also on: Lon Harris, editorial director at LAUNCH
Published: 28 August 2026 on This Week in Startups
Episode page | 1 hr 48 min
Key Takeaways
Nvidia buying Hugging Face is the year's biggest story, not the blowout quarter
"They've agreed to acquire Hugging Face for 12.9 billion." — Lon Harris
Jason's read: America had no open-source champion, and now it is Nvidia
Unmetered on-premise racks are what should scare OpenAI and Anthropic
"Unmetered is the word we're going to be talking about in 2027." — Jason Calacanis
He expects it to land in both companies' S-1s as the number one risk factor
Cheap, relentless intelligence makes hacking effectively unbounded
Jason expects a cataclysmic hack before the end of the year and reads the open letter as pre-positioning
Palo Alto Networks and CrowdStrike are the names Jason says benefit
Halter gives the collars away and charges by the cow-month
A base tier of about $9 per cow per month, with roughly one and a half million collars in the field
Starlink direct-to-cell is what made virtual fencing work in mountain country
The collars mesh together and elect a leader to talk to the satellite
Five years of hardware work came before Halter had anything to sell
Ranch software is sold face to face, not over Zoom
About 250 field reps out of roughly 600 employees, with Toast as the explicit model
Drill straight down before you go adjacent
Jason's Uber example: same product, new geographies, and no Uber for boats
Hop Aero is seven people building a 550-pound cargo rocket
Matija Milenovic says AI-assisted engineering lets them move like a team of 50 to 70
Hardware in the room beats renders when you are selling to the Pentagon
An AI-written op-ed under a human byline breaks the promise the byline makes
Lon Harris: the reader was told the argument was the author's own work
AI's Oppenheimer Moment: 100 Companies Warn About Cyberattacks
Lon Harris opened with the open letter: more than 100 companies, including OpenAI, Anthropic, Google and Microsoft, published a call for collective action on cyber defense, warning that "we only have months to prepare for AI enabled cyber attacks"
Hospitals, water systems and critical infrastructure are named as exposed
The signatories want raised security standards, shared threat intelligence and funded defense at every level
Lon's own caveat on the letter: "the letter doesn't have any actual commitments." No deadlines, no mandated funding, just a suggestion floated at politicians and the public
Jason turned it into a live lesson in what he calls media deconstruction: ask why now, ask who the letter is addressed to, then check whether the individual leaders are actually posting about it
The letter sits on OpenAI's site and ranks first in search, which he read as OpenAI leading the pack
Lon checked the accounts on air: Sam Altman had not tweeted since 21 August; Dario Amodei had not posted since what Lon called the Gavin Baker kerfuffle of the week before; Satya Nadella was silent on it
The one leader who did speak: Altman, on 27 August — "This is a critically important moment for cyber defense with AI. There is not much time to act." Jason called the framing very charged and very dramatic
Jason's explanation for the timing is an incident nobody has announced yet. He pointed to OpenAI's 26 August post about an investigation into the Hugging Face incident, in which its own agent went after credentials on Hugging Face's site
His inference: if that happened once, other agents may be hitting other people's sites during model builds
"Always when these letters come out, always when people make statements, there's usually an incident that people know about"
He predicted a significant hack of government systems or personal data landing within days, and framed the letter as the industry warming the public up — "we've already probably been hacked by AI"
Jason also contrasted this with what he described as China's PR campaign preparing its own population for robotics: "They are prepping them to accept robotic companions, robotic workers"
Lon's read on the letter's tone: "it is a little blameshifty." His question back: is this all of our problem, or is it the signatories' problem?
Why Frontier Coding Models Make Hacking Unbounded
Jason's core claim: a model good enough to write code recursively is by definition a model good enough to hack at infinite scale. The counter-tool to a great vibe-coding model is a tireless attacker limited only by available compute
Lon on the asymmetry: "There's no one human who can put all of GitHub, every open source code into their brain at once."
Jason walked through what such a system would actually do: rank every open-source project by stars, update frequency and neglect, cross it against the history of every hack since 1970, and find the soft targets
"it can guess your next word, it can guess the next hack"
Lon added the cost curve. Attackers are budget-constrained and cannot run the most expensive models around the clock, but they can already afford cheaper ones like DeepSeek, and those keep getting both cheaper and more capable
Jason's trade off the back of it: "this is going to be a boom for Palo Alto Networks. It's going to be incredible for CrowdStrike." He called them the two leaders and said he would need to do a proper analysis of both
His timeline: "I think we're going to see a hack before the end of the year." The scenario he sketched is an at-scale phishing attack that catches 2% of people, then 10%, followed by drained crypto wallets and dumped private messages, with extortion behind it
He argued that would turn public feeling against AI far harder than data centers or AI slop have
The analogy he expects to be used against the labs later: cars sold without airbags, cigarettes sold as non-addictive. He cued up the 1994 congressional hearing where tobacco executives denied nicotine is addictive, and predicted Altman and Amodei on a stand in about 18 months
"you gave people a car you left out the airbags and the three-point seat belts"
Both hosts recommended Michael Mann's 1999 film The Insider off the back of the clip
Halter's Collar: Virtual Fencing Sold by the Cow-Month
Craig Piggott described Halter as infrastructure rather than a gadget: "It's like a tech platform for ranchers", with a solar collar at the center of it
"we use that collar to train animals to respond to cues", which lets ranchers fence and move cattle without physical fences
The same collar monitors health, fertility and calving, and the platform measures grass
Jason's description on seeing it: "It looks like a Fitbit with solar panels for cattle."
The core sensor is an inertial measurement unit. Craig: "The core sensor that powers a lot of the analytics is the IMU." Temperature and humidity are tracked too, but health, fertility and calving are inferred from behavior using models run on the collar
The training cue is audio, not shock. Craig's analogy: "The best analogy is when you're parking your car" and the sensors beep as you approach a wall
Cattle are trained with a very low energy pulse that gives the sound meaning, after which it runs on sound and vibration
Most cattle are already trained to an electric fence, where the primary cue is vision; Halter turns that into audio
The business model is a subscription and the hardware is free. Craig: "You don't buy the hardware. We give the hardware away for free. You just pay the subscription."
"The like base starting tier is around $9 per cow month", with multiple tiers and packages depending on the operation
Jason did the arithmetic out loud: "So you can spend 200 bucks a year per cow. They're paying $200 a year."
Scale to date: "we've got about one and a half million collars out there" across New Zealand, Australia, the US and Canada
Collars move between animals — scanned with a phone and reassigned — and beef operations tend to collar the mothers rather than the calves being raised for the end product
The headline value proposition is land, not livestock. Craig: "fences have given ranchers control of their land for hundreds of years, but they're really expensive" and cannot be moved
"we've got some customers that are running twice as many cattle on the same piece of land" because they can graze the spots cattle otherwise avoid and rest the ground when it needs it
Getting Data Off a Cow in the Middle of Montana
Jason pressed on connectivity, since ranch country has neither power nor cell coverage. The old answer was base stations and towers, which worked on flatter country and failed in the mountains of Montana, Wyoming and Colorado
Craig's answer is satellite: "each collar now can talk to Starlink and they also talk to each other." The collars form a mesh and "they elect the best leader to then talk to Starlink", which saves power and cost
The radio is the same direct-to-cell technology a phone uses — Craig said the collar is "It's about the same size as say a smartphone on each side of the cow"
Throughput is thin, and Halter designs around it. Data is coordinated across the herd, compressed, and sent in bursts: "it takes like a few seconds every minute to transfer that data"
"We do run a lot of machine learning models on the device itself" — where 100 Hz sensor access is needed, the model is trained offline and pushed onto the collar, which removes the bandwidth problem entirely
What Ranchers Actually Want Once They Live in the App
Lon, who had called the product Whoop cows in the pre-interview, asked what a rancher actually looks at. Craig's answer: the monitoring is a small part next to fencing and shifting, but health and fertility matter most
Animals can be hours away, and a rancher might spend four hours on horseback looking for one
Halter's roadmap is customer-driven, because customers live in the product. Craig: "They log in 50 times a day." The philosophy is the best engineers, data scientists and designers, then let customers drive the feature list
Grass measurement and grazing recommendations came out of customer requests
So did sickness alerts — customers could see it in the data themselves: "she's clearly sick. She's grazing less. She's walking less."
Devices are updated every couple of weeks, which Craig said ranchers have never had before
No cameras on the collars, but plenty of remote sensing. Craig: "We measure what they're eating, how much grass there is, like all of that information through using satellites", combined with weather stations and collar data
"we don't have cameras on the device" — cattle are not in one fixed place, so a camera would miss what you need; the app does let ranchers photograph their land for an assessment
Coming next: "measuring intake through the collar by counting bites" and profiling them, so cows can be ranked on efficiency
Jason's own analogy came from his wrist. He said his Whoop has alerted him twice in two years on skin temperature, and both times it preceded a cold
"if everybody had whoop during the pandemic", he argued, people could have been told 48 hours earlier and treated inside the window when antivirals work
He put the same logic on cattle: catch the animal before it gets sick. Craig priced the animal at "three and a half grand at the moment"
Craig's example of how bad the status quo is: on public forest service land, "they'll turn out 500 cattle and they go to gather them and they spend three weeks up there with 10 cowboys" and come back missing 30 head
Jason also told a ranch story of his own: two of a neighbor's cattle came through a downed fence, his wife suggested they were now theirs, and he explained that cattle rustling is the one thing you do not do in Texas. The neighborly protocol is to walk them back and rebuild the fence together
Five Years in the Laboratory, and Parents Who Thought It Was a Fun Little Thing
Product-market fit took longer than two years, and the constraint was hardware. Craig worked at Rocket Lab before starting Halter — "before Halter I was working at Rocket Lab, grew up running cattle" — and described the early period bluntly: "that was a tough tough five years in the laboratory"
The problems were keeping collars alive through winter, talking to them, building the hardware and training cattle
All of it was tested in the field on his parents' ranch: "they've helped us so much by just giving us free rein of their" land and cattle, and "we've been buying and selling trying to work out how to train different breeds"
His parents were supportive but skeptical. Craig's summary of their early read: "Craig's just doing a fun little thing"
Jason turned that into the founder lesson of the segment. Resistance comes from good people, because "the more time you've spent in a vertical the codified your belief system becomes"
"change is hard for people", which is why outsiders reinvent hotels, delivery and stock trading — his examples were Airbnb, Uber and Robinhood
The founder's job is to walk incumbents through the vision step by step
Craig's counter-framing: coming back to farming after engineering school and Rocket Lab helped, but "if you build a product that's so good, they can't ignore it" is what he keeps returning to
The Rocket Lab detour produced a brief tangent on its founder, Sir Peter Beck, and whether a New Zealand knighthood is the same as a British one. Craig: "We haven't had our independence moment."
The Sales Motion: 250 Reps on the Ground, With Toast as the Model
Craig on how Halter sells: "The sales motion is outbound field-based hyper local." A rep sits in each community and is expected to know every vet, consultant and rancher in it, effectively as a territory manager
The model was borrowed openly. "Toast became our inspiration" — restaurant owners do not think about point of sale, ranchers do not think about collars, so you have to get in front of them and show them
What actually drives growth is neighbors talking: "word of mouth is it's the primary driver." The product does what it says on the tin, customers love it, and they tell the ranch next door
The staffing ratio is the number Jason flagged as the lesson. Craig: "we probably have 250 reps"; Jason: "600 people at the company. A third of them are doing the go-to market."
Jason's framing for other founders: this is not an enterprise sale you can run on two 20-minute Zoom calls. Ranchers, farmers and vets have to trust you first
On the business itself, Craig said: "We're about 600 people. Growth is really strong." The US launched about two years ago, South America is next, and the company is hiring in every market
Reddit Thinks the Collars Are for People
Lon put the internet's conspiracy theory to Craig directly: whenever Halter comes up on Reddit, "people are very worried you're going to put them in the collars"
Craig's answer covered both the joke and the strategy: "We're doing nothing but cattle for a long long time." The company is asked daily about goats, sheep and horses and says no to all of them
He added that the theories were not helped by the company's recent Founders Fund round
Jason's take on why that round attracted it: anything touching Peter Thiel becomes conspiracy corner. "It is conspiracy corner for sure anytime" his name is attached
Jason also joked that he had arranged a collar for Lon, whose performance was low and who had been grazing too much, and that it would be on his desk after the show
Jason's serious defense of the category was conservation: more efficient use of land, less disease and a better food supply. He said owning a ranch changes how you think about working the land, and that he will likely rent part of his to cattle ranchers
Craig's pitch to engineers is that the work is physical and the customers are desperate for it. Ranchers, he said, "they some of them will cry like in a salesman about the potential" because they have had no tools at all
Drill Straight Down Before You Go Adjacent
Jason's tactical lesson of the episode is against founder ADHD. When you strike oil, keep drilling in the same hole
"Step one when you hit oil is to keep going straight down." For Halter that means cattle in New Zealand, cattle in Australia, cattle in the US
Horses are the trap: owners keep three, not a thousand, they are pets rather than food, and the adjacency means starting over
His model is Uber under Travis Kalanick — same product into Vegas, London and Australia rather than Uber for boats or planes, with Uber Eats only once the core was pumping
Craig agreed and gave the sequencing rule: "I think you want to do things in series, not parallel when you can."
Halter was exclusively New Zealand and growing "3x year on year for many years before we even thought about Australia" and then the US
Even inside one country the runway is long: "there's 10 million cattle in New Zealand" and Halter is on maybe 5% of them
On the size of the prize, Craig said: "There is about 1.5 billion cattle in the world", which makes the current install base "a very depressingly small percentage of penetration"
Jason's investor rebuttal: "That's actually called an opportunity in my business." The founder sees how little is done; the investor sees the market. "You literally can go a 1000x from here."
Craig pushed back gently on the pure-TAM read: not every animal is in the sweet spot, and Halter increasingly sells product that sits apart from the collar — "we view ourselves more as like also a tech platform for this industry", including satellite products and agents that help run farms
Jason's other Uber lesson was writing the playbook down. Kalanick had different people build the New York and California playbooks, then compared them: driver recruitment transferred, delivery courier recruitment did not, because New York runs on bikes and California on cars
On US scale, Craig said the biggest customers are enormous — "some of them are probably bigger than half of New Zealand as a country" — and that the Starlink product is what makes them reachable
The Four People Jason Says He Owes His Career To
Jason used the gap between guests for a long personal digression about how he got into media, prompted by his guest appearance on No Agenda — "I'm going to guest host No Agenda on Sunday", live at 12:30 Texas time
The first name is John C. Dvorak, who died last month. Jason described tearing open PC Magazine at 13 to get to Dvorak's column, and the ambition it produced: "Someday I would love to get paid to write about computers."
"John was my inspiration as like a personality" — spicy, opinionated, and worth paying attention to, alongside Howard Stern on the radio
He noted the ages plainly: "I'm 55. Adam's like 61. I think John made it to 75."
The second is Dave Winer, who Jason credits with HyperCard, RSS and OPML — "he created HyperCard. He created RSS and OPML"
The third is Adam Curry, the MTV VJ who asked Winer to add an enclosure to the RSS spec and then started attaching MP3s, which is how podcasting started. Curry's domain buying also got Jason into premium domains
The fourth is Leo Laporte, whose This Week in Tech was the direct model for this show. Jason asked permission and was told the this week in branding belonged to nobody
"I have those four people to thank for a large portion of my career."
RSS is what pushed Jason into blogging. Seeing that readers could subscribe rather than visit, he called Brian Alvey and started Weblogs Inc: "I'm going to make a hundred blogs" at a time when only Nick Denton had more than one — "Nick Denton had three", which Lon corrected to two, Gawker and Gizmodo
Lon's framing of why RSS mattered: "It's hard to explain how revolutionary RSS was to people today." Before it, if the New York Times would not publish you, you were not published
Jason asked Lon to cut MP3 clips for the Sunday appearance, since No Agenda is audio-only and built on playing clips and responding — "I'm going to need to work with you on Sunday morning."
Hop Aero's Rook: 550 Pounds, 450 Miles, 15 Minutes
Matija Milenovic introduced Hop Aero as point-to-point cargo delivery by rocket, aimed at time-sensitive or high-criticality assets anywhere on Earth in minutes
The first vehicle, Rook, has "a range of on the order of hundreds of miles that can do that in about 15 minutes"
Dimensions are not public, but "we're not publicly disclosing the exact dimensions of Rook, but very roughly 20 foot long"
"we're planning on having a payload capacity of about 550 pounds"
The advertised use case is moving autonomous systems around the Indo-Pacific — one point to another in 15 minutes, "that's about 450 mi away"
Payloads named: "deploy swarms of autonomous drones or counter UAVs", electronic warfare and spectrum jamming, local compute and local power
Matija's argument for why this matters is about who is on the front line. Since the beginning of time, "it's been we've sent out 18-year-old boys into the front line" with spears, then bullets, then worse
The trend is "getting these 18-year-old boys out of harm's way"
"whoever is able to deploy autonomous capacity at range and at scale is ultimately going to be able to deter the next major conflict", and he called Ukraine a small-scale version of the experiment — "it's been shown in Ukraine in a kind of small scale"
Jason opened the segment on the consumer version of the same idea, relaying Elon Musk's pitch for rocket travel between continents: "We can get to Japan in like two hours", at roughly $150,000 round trip
Make It Real: How Seven People Get Taken Seriously by the Pentagon
The company is under two years old and in the current Y Combinator batch. Matija corrected Jason's assumption of three or four years — it is just under two — and said "We're still in the current YC batch." The amount raised is not public
The why-now is AI reaching hardware engineering. Matija: "AI is able to augment a lot of the engineering development", so it is seeping into hardware, not just software
"we're a very small team of seven, and we've been able to move at the speed and ferociousness of a team of probably 50 or 70"
He credited the path being open at all to SpaceX and to defense companies like Anduril: we all stand on the shoulders of giants
Jason offered unsolicited fundraising math for the audience, stressing it was a guess: "you need $30 million in this like series A", or 20 to 30 million, at a 100 or 200 million valuation, diluting about 15%
What actually moves a defense conversation is hardware, not slides. Matija: "renders and powerpoints and all of that kind of takes you so far", but hot-firing an engine you built from scratch in a couple of weeks speaks volumes
"It's actually showing clear tangible progress" rather than prototypes alone
Concept conversations warm things up — "you talk about an OV1 in the military" — but "you just got to whip out the iPad and start showing actual footage", or better, host them at the test site: "we've had guests that we hosted live hot fires for it"
Jason named this the tactical takeaway of the segment: make it real. Anyone can make a pitch deck; assembling a team builds credibility; firing an engine without blowing yourself up is zero to one from deck to real capability
There is a commercial business behind the defense one. Matija said Hop Aero has signed early pilot customers "ranging from manufacturing to cold chain, transportation, automotive"
The economics are downtime, not cargo value: "It's what's the cost of not having it."
Even so, "we've had far more pull from the Department of Defense or Department of War", so that is where the energy goes
On the department's renaming, Jason took a side: he prefers Department of Defense, called the war framing too aggressive and like picking a fight, and predicted it changes back in two years
He also credited the current administration for telling startups it wants lower costs and early contracts — "this administration for the first time has said startup, startup, startups, lower our cost"
Lon tried to bait him into criticizing Pete Hegseth; Jason declined while a guest with a deal to close was on the line
Hiring is across GNC, aerothermal, structural and assembly roles, to be posted on the company's site and LinkedIn
Nvidia Buys Hugging Face and America Gets an Open-Source Champion
The deal, as Lon laid it out: "They've agreed to acquire Hugging Face for 12.9 billion." Hugging Face is "the public marketplace for open source AI models data sets the GitHub of machine learning"
"Nvidia tried to buy it in late 2025 but they were turned down"
The earnings context Lon read out: Nvidia's report "forecast 70% 2028 revenue growth while analysts were expecting just 44%"
Jason's read is that Jensen Huang is open-maxing on purpose. Nvidia already runs open-source work in Nemotron and an open self-driving stack with Uber, and Jason believes Huang genuinely believes in open source
But the customer base is frontier labs: he named OpenAI, Anthropic and Google as the companies behind the record quarter, and added Microsoft to the list
The strategic gap Jason says Huang spotted: "There's no champion. There's no Anthropic of open source yet." Lon's attempt at an answer was Meta, and Jason's response was that nobody is in the lead
"US open source equals Nvidia" is the sentence Jason says people will be saying
"Chinese open source equals Kimi", with Lon adding DeepSeek and Moonshot to the list of labs on that side
Jason called it the news of the week over the quarter itself: "This is the biggest story of the year." And, he added, nobody is talking about it
How Sam and Jensen Went From $100 Billion to an Open Relationship
Lon walked the timeline. On 22 September 2025, "Altman did a huge hundred billion dollar deal with Nvidia" for chips and compute — the deal Jason says started the whole circular-revenue conversation
Fourteen days later Altman announced an equity agreement with AMD giving it "warrants to buy up to 10% of OpenAI", another circular deal
By January, per the Wall Street Journal, Huang was telling industry colleagues that "he was already telling industry colleagues that his letter of intent with OpenAI is non-binding" and criticizing a "lack of discipline in Altman and OpenAI's approach"
"Altman then brought in Google vet Richard Ho" to build OpenAI's own inference chip, now called Jalapeno
Jason's summary of the relationship: it went from a hundred billion dollars in commitments to an open one, and Huang responded by acquiring inference capability of his own — "Of course, Jensen buys Groq, gets his own inference chip. Cerebras goes public with their amazing inference chip."
The other deal of the week is Poolside. Lon: Nvidia is taking around "They're going to take like a hundred engineers from poolside" in a structure of "$6 billion plus a billion dollar or $1 billion deal plus a licensing fee of $6 billion", absorbing one of the last large independent coding labs
Poolside's coding product is called Laguna, a competitor to Cursor, Codex and Claude Code
Lon asked which brand survives, Laguna or Nemotron; Jason's bet is Laguna, under Nvidia
Unmetered Is the Word for 2027
Jason's central thesis is that Nvidia now sells the whole stack into its own customers' customers. "Jensen is going to take open source. He's going to roll a rack into companies."
The pitch to a CIO already buying frontier models: "We power 72% of what they do", and then, "Might I interest you in a rack that has a Codex competitor that is unmetered?"
The worked example is Uber. Jason imagined Huang asking what the company spends on tokens, guessing four to five hundred million, and offering to take a hundred million of it in racks that ship tomorrow
The sweeteners are on-premise data that never reaches a frontier model, AI sovereignty, and no surprise invoice: "There's no shocking bill at the end of the month."
The CFO's version of the appeal is a capped number — a hundred million in servers and electricity — instead of an argument about AI spend
The claim that gives the segment its name: "Unmetered is the word we're going to be talking about in 2027." Not token-maxing, unmetered
Jason thinks this hits the labs where they earn: "The frontier model coding products are going to have a really hard time competing with an unmetered option."
"This is going to be in the S1 for Anthropic and for OpenAI as their number one risk factor." Not data centers, not cybersecurity
"That's why he bought Hugging Face. That's where the developers hang out. That's why he bought Poolside." — and with Laguna, "he can now go after the two most profitable parts of OpenAI and Anthropic."
The bundling argument is about vendor count. A startup selling software to run on someone else's rack is two or three vendors; Nvidia is one — "It's going to be one vendor coming in" with the whole stack
It also solves the will-this-company-exist problem: "the largest company in the world with $5 trillion has made a $20 billion commitment to open source"
The size of the prize as Jason frames it, conditional on it working: a third of the top two frontier labs' market capitalization — "That's $1.5 trillion in additional market cap moving from Anthropic" and OpenAI to Nvidia
A Detour Through Laguna Beach
The Poolside product name sent the hosts into a tangent about Laguna Beach. Lon's history: "It used to be like in the ' 60s7s it was very bohemian", an artist colony with a funky vibe distinct from conservative Orange County
His view of it now: "I feel like less so today. It's become gentrified like everywhere else in that part of Orange County", closer to Huntington Beach or Newport Beach
Jason's parallel was Los Angeles: "Venice was like a weird art community. Now it's like a yepy community." Lon's explanation is cost — the hippies and bohemians got priced out
Both agreed Laguna is worth a night on a drive to San Diego, and Lon's advice was "Don't stay for more than 48 hours. Gets a little weird."
The Druckenmiller Op-Ed and the Promise a Byline Makes
The subject is the Wall Street Journal op-ed published under Stanley Druckenmiller's byline that appears to have been written by AI. Jason asked Lon, as a writer, what a reader is entitled to expect from a premier paper's editorial page
Lon's argument is about the choice of venue. A billionaire has a million ways to communicate — call a reporter, hold a press conference, post, film himself, go on a podcast. Choosing an op-ed says something specific
His comparison was pamphleteering: Hamilton and Thomas Paine putting an argument into the national conversation
"To me, that's a promise to your reader that writing is your own", that it is your argument, carefully advanced, every sentence original
Lon disclosed his own newsroom's policy at LAUNCH, where he is editorial director: "we don't publish anything front-facing that is just written by AI." They use AI constantly — ideation, research, drafting
"I think the hardest thing about writing personally is the blank page." AI gets something down; "It's much easier to edit than to write."
The essential part is the editing pass that makes it his own work, which is also where the thinking happens
Jason's position is a disclosure rule, not a ban: "if you're going to write something with AI, it should be labeled as such in the first sentence up front"
His analogy is documentary practice — "if you make an AI video of a shark attack and it's a recreation of it", you label it a recreation, because the audience expects real footage
He also reached for Milli Vanilli and for a friend whose documentary appearance was altered: "They used ElevenLabs to put in a sentence he didn't say"
"I feel like you've broken the trust here. Just like Milli Vanilli broke the trust."
His verdict is on both parties: "the mistake that the Wall Street Journal made and then subsequently Druckenmiller made" was not checking the work and not disclosing it — "Wall Street Journal failed at their job. Druckenmiller failed at his job."
"If you've eroded the trust of the reader, the consumer of the information, you failed at your job."
Lon's point about the byline itself is the sharpest line in the segment: "There's a by line that says written by Stan Druckenmiller right there at the top. That's a lie." If he had co-written it, both names would be on it
He checked for a disclosure after the controversy: "I revisited it this morning and did not see one."
The paper's position, per Lon, is that "this represents Stanley Druckenmiller's point of view" and he approved it for publication
Lon's dividing line is human in the loop: "It's easy to understand the distinction between no human in the loop and human in the loop." Grammarly and search adapt your writing; a model composing it is doing the writing
Jason expects the scandal to widen, and to become political. The tweet that surfaced it came from Republican strategist Andrew Surabian — "This is a Republican strategist with 97,000 followers" whose avatar is Roddy Piper from They Live — alleging an editorial "attacking JD Vance was 100% written by an" AI
Jason's prediction is a mass audit using the detector Pangram: "we're going to find out who's not wearing shorts cuz the tide's about to go out"
The pair went down a lawsuit tangent — the early-1990s class actions over lip-syncing at pop concerts, with Lon reading a Chicago Tribune piece on New Kids on the Block fans suing in Cook County. Jason did not think this case needs a lawsuit
A brief aside on Jason's own reach: "Did you know I have 1.7 million followers on X?" He said he thought he had just crossed a million, and Lon put the start of 2026 at about 1.2 million
Off Duty: A Laser Iron Dome for Mosquitoes
Lon raised the mosquito problem in Austin and the blue-laser insect defense system Jason had tweeted about, comparing it to an iron dome for a backyard
"It uses sensors to find mosquitoes midair and zap them." It "It scans a 20 foot 90 degree field with lidar, radar, and AI vision. It can target insects as small as 2 millimeters" and "runs for four or five hours on one charge"
Jason's own solution is cheaper and already installed. He described "This is a $250 device meant for like barns and places where you have animals" and said "These things supposedly do two acres."
He has two big units outside and smaller ones inside, and says his family went from five to ten bites a week each to none
The reason he stopped debating the ethics of killing mosquitoes was a skin infection. Scratched bites got infected and he needed a prescription cream — "I had to get mupirocin from an emergency care clinic" — and started wearing long pants to stop scratching
Off Duty: The Wizard of the Kremlin, King and Conqueror and Lanterns
Lon's first recommendation is a Putin film: "It's the wizard of the Kremlin it's called." A fictionalized account of the rise from KGB officer and bureaucrat to entrenched autocrat
"Jude Law stars as Putin. Paul Dano plays his propaganda minister." — "That's on Hulu and Disney Plus right now."
The director also made Carlos, which both hosts rate: "we've talked about how much we like that Carlos the Jackal movie". Lon was unsure of the pronunciation on air — "Asas or Asaas. I'll look up how to pronounce it."
Jason's addendum: the Carlos miniseries runs five to seven hours against a three-hour cut, and he believes the full three-part version is on Criterion Channel
The second is a UK historical drama on Prime Video: "It's called King and Conqueror", about "the Norman invasion of England and the Battle of Hastings in 1066", with Nikolaj Coster-Waldau and James Norton as Harold II and William the Conqueror
On Lanterns, both are watching and both call it slow. Lon likes that it feels like its own show while staying faithful, the way Andor and The Penguin do, and rates Aaron Pierre as John Stewart
"It's very evocative of Ozark, which was a great series." Jason's comparisons were True Detective, Ozark and Taylor Sheridan's Landman
Jason's verdict: "it's a slow burn" and better binged over four or five days than watched weekly
The Aaron Pierre film they landed on is Rebel Ridge on Netflix — "Rebel Ridge is a great popcorn movie", with Don Johnson as the sheriff
The $19 Uniqlo Bag That Beat the $160 One
Jason had previously recommended the ALPAKA Bravo Sling Max 8L — an "RFID blocking sling laptop bag I bought for 160 bucks" that "it fits my MacBook Pro 14-inch", which he considered a reasonable price and a smart buy
Then he found a near-identical Uniqlo shoulder bag in Tokyo. "Here's my Uniqlo bag that I bought in Japan for $19." It is "29 bucks in the US"
What it loses: no Faraday pocket, no pass-through USB charger. Otherwise he says it is essentially the same bag, slightly lighter, and he reaches for both equally
He ordered six for the team on air
The bag became a Uniqlo pitch: "I want to be on the board of Uniqlo to learn." He put it out into the universe that if the company ever wants an American director, he is available, partly for the excuse to go to Japan
His stated reason for caring at his level of wealth is temperament: he loves value, and finding something that one-ups a good value is the part he enjoys
Most Value for the Least Amount of Work
Jason closed by making his editorial director's assignment public as a teaching exercise. "Lon is been given the directive to make this week in startups, this week in AI, and this week in VC. Substacks."
The AI newsletter already exists at thisweekinai.ai and, per Jason, "It's got 8,000 subs." This Week in VC has not been set up yet
He said he pulled Lon off other work because Substack is its own ecosystem and the show was not paying enough attention to it
The design question he wants answered: what is the least amount of work that produces the maximum amount of value?
The precedent is Hackaday, the most profitable blog at Weblogs Inc. It ran one item a day: "we featured somebody else's hack with one sentence describing the hack and a link to it"
"It did one thing every day", and the work was the curation — a couple of hours to find the hack, ten minutes to write the sentence
Jason's regret is that he sold it. When AOL bought Weblogs Inc, legal wanted 17 Hackaday posts taken down because "you're teaching people how to hack the Time Warner" set-top boxes
His counter was that the site was de minimis — "this thing makes like $300,000" a year against "It's costing us $500 a week" — so he asked them to take it out of the deal
"I took some of my proceeds from the deal and I bought it." "Then I wound up selling it for a million dollars or so", netting a couple of hundred thousand, and he says he should never have done it
The assignment has a deadline: "I want you to report back here on your plans for those and how it's going", next Friday, so the audience can watch the editorial trick get applied
He also asked for a dedicated email around the first how-to short, sent with links to where the videos live, and for a count of how many people subscribe off it — plus a question to readers about which how-tos they want next
Jason's bottom line is that the week's real story is not Nvidia's quarter but Jensen Huang buying his way into open-source coding tools, because an unmetered rack sitting inside a customer's building is the one product the frontier labs cannot price against.
Products, Companies & Tools Mentioned
Nvidia, Hugging Face and Poolside (The week's three-way story: a $12.9 billion acquisition of the open-source model hub, plus roughly a hundred Poolside engineers and its Laguna coding product, which Jason says lets Huang attack the most profitable parts of OpenAI and Anthropic)
OpenAI and Anthropic (Nvidia's two largest frontier customers and, in Jason's telling, the two companies whose coding revenue an unmetered rack threatens)
Halter (Craig Piggott's virtual-fencing platform: free solar collars, a subscription from about $9 per cow-month, roughly one and a half million collars out, about 600 people)
Starlink (Direct-to-cell connectivity that made Halter work in mountain country; collars mesh and elect a leader to talk to the satellite)
Hop Aero (Matija Milenovic's cargo-rocket startup; Rook carries about 550 pounds roughly 450 miles in 15 minutes, built by seven people in the current YC batch)
Palo Alto Networks and CrowdStrike (The two security leaders Jason expects to boom as frontier models make hacking cheap and relentless)
Whoop (Jason's own wearable, and the analogy for what Halter's collar does for cattle — early warning from skin temperature)
Toast (The outbound, hyper-local sales model Halter copied for selling to ranchers)
Uber (Both the focus lesson — drill straight down, no Uber for boats — and the hypothetical customer for an unmetered Nvidia rack)
Rocket Lab (Where Craig worked before Halter; its founder Sir Peter Beck came up in the tangent about New Zealand knighthoods)
AMD, Groq and Cerebras (The inference chess board: AMD's warrant deal with OpenAI, Nvidia's Groq acquisition, Cerebras going public)
Jalapeno (OpenAI's in-house inference chip, built under Richard Ho after the Nvidia relationship cooled)
DeepSeek, Kimi and Moonshot (The Chinese open-source labs Jason says have no US counterpart — and, per Lon, the cheap models attackers can already afford)
Founders Fund (Led Halter's recent round, which Craig says fed the online conspiracy theories)
Y Combinator (Hop Aero is in the current batch)
Anduril and SpaceX (The defense and space precedents Matija credits for making his company fundable at all)
Pangram (The AI-text detector Jason expects will be run across a year of newspaper editorials)
ElevenLabs (Used, per Jason, to put a sentence a friend never said into a documentary)
Uniqlo and ALPAKA (A $19 Tokyo shoulder bag against a $160 RFID-blocking sling; Jason wants a Uniqlo board seat)
Weblogs Inc, Hackaday and AOL (Jason's blog network, its most profitable site, and the acquirer whose lawyers wanted 17 set-top-box hacking posts removed)
Substack (Where Jason has directed Lon to build three show newsletters)
Books & Resources Mentioned
The Insider – Michael Mann, 1999 (Cued up alongside the tobacco hearing clip; Lon's line was that if you haven't seen it, watch it)
No Agenda (Adam Curry's podcast, which Jason guest hosts on Sunday)
thisweekinai.ai (The show's AI newsletter, which Jason says has 8,000 subscribers)
The Wizard of the Kremlin (Jude Law as Putin, Paul Dano as his propaganda minister, on Hulu and Disney Plus)
Carlos (The same director's earlier work; the full miniseries runs five to seven hours, Lon believes on Criterion Channel)
King and Conqueror (Prime Video series on the Norman invasion and the Battle of Hastings in 1066)
Lanterns (The DC series both hosts are watching; slow-burn, with Aaron Pierre as John Stewart)
Rebel Ridge (The Aaron Pierre film on Netflix that Lon calls a great popcorn movie)
Get the latest market chatter and takes as they happen:
X | Threads | Instagram | YouTube | TikTok | Facebook