David Weisburd 🚀 retweeted
Crazy idea: Let's split a country in socialist and capitalist halves and check in on them in 75 years:
Ethan Thornton (@ethanrthornton), founder of Mach Industries, on how fast Pentagon procurement has changed:
"When I first pitched one of our products to the Army, they said, this is incredible, we'll roll it into our 2035 plan. Now they show up and say, we need you to build something. What can help?"
No more decade-long roadmaps. Align on what's needed, then move out.
He's blunt about why: Iran burned through a significant chunk of its munitions in a few weeks. If it takes the US years to rebuild stockpiles after a real conflict, that's not a budget problem, that's a national security problem.
Speed and collaboration with startups just became non-optional.
John Borthwick (@Borthwick) has his AI agent write a blog post after every project on what it learned, not just about the task, but about him.
"I learned that John has a completionist tendency. Getting to 100% is hard, but 80, 90% is easy."
The point isn't the blog post. It's that you're not spinning up a new instance each time. You're training the same agent on you, over and over.
He thinks the memory layer is the whole game. Important enough that he's backing a company, Plastic Labs, just to build it.
John Borthwick (@Borthwick) on Elon's idea for peer to peer LLM testing on security:
Every model has its own style, its own way of solving problems. Point them at each other's security holes and you're attacking the same vulnerability from a dozen different angles instead of just training a model against itself.
His caveat: it has to sit outside the labs. After the Hugging Face and OpenAI incidents, most of the good research came from alignment teams. But alignment and security aren't the same problem, and we haven't solved either one yet.
China's procurement system moves faster than ours. Their industrial base is bigger. They have more drones deployed in Ukraine than we do.
Ethan Thornton's (@ethanrthornton) point: counting on mutually assured destruction when your conventional weapons are behind is how you get proxy wars, not peace. The Cold War was a story of proxy wars in Korea, in Vietnam, everywhere deterrence didn't quite reach.
His fix isn't more nukes. It's arming Taiwan, Japan, South Korea, Australia with asymmetric drone systems so tactical deterrence works the way strategic deterrence has for decades.
His read on Russia: if they understood unmanned systems the way they do now, they don't invade Ukraine.
3
424
Betaworks ran its first agent camp two and a half years ago. Early, by most standards.
John Borthwick (@Borthwick) on why that timing mattered when OpenClaw showed up last December:
"We see this. We see you. We're very interested in this, and we kind of get where this is because we have that depth of experience."
The camps are time gated. Great companies aren't. So Betaworks built the ability to write checks outside the camp cycle entirely, on conviction from a thesis they'd already gone deep on.
Depth built two years ago is what let them move in days, not quarters.
David Weisburd 🚀 retweeted
BREAKING: Jensen Huang reportedly painted a fake tunnel on a brick wall, tricking an employee into running headfirst into it.
John Borthwick (@Borthwick), on what's actually happening inside the labs right now:
"We've basically taken sand, silicon, and a lot of power, and we have found some form of intelligence in there."
Research labs that discovered something new are now scrambling to become product companies. That mismatch, he argues, is exactly why the people closest to the technology keep reacting with "holy shit."
Because it is holy shit.
"Cost plus contracting is another form of socialism, and that's as clear as I can put it."
Ethan Thornton (@ethanrthornton) on why decades of cost-plus defense contracts breed companies that have to overcharge the Pentagon to survive.
His point: incentives drive outcomes. Cost-plus punishes efficiency and rewards bloat. Fixed-price forces companies to actually compete on cost, capability, and reliability.
"You need to be deeply capitalistic to ship great technology."
That's exactly why SpaceX won.
Ethan Thornton (@ethanrthornton), founder of Mach Industries, on what Ukraine actually proves about asymmetric warfare:
Not bigger industrial capacity. Not better engineering. Staying on the Pareto edge of technology as the fight evolves.
"Asymmetry early in the war was them adopting traditional Western Javelins incredibly well... but asymmetry is a fleeting thing. The tech on the front edge of that curve changes all the time, and so you have to actually innovate to push super hard."
Right now that edge is deep strike. Ukraine picking off Russian infrastructure, Russia doing it right back.
Whoever innovates next defines the next war.
5
719
AI finds vulnerabilities by the million, but just one in 1.3 million is exploitable, says cyber CEO Sanaz Yashar. Her fear is that attackers chain the weak ones.
@sanazyashar is Co-Founder and CEO of Zafran Security. On the discoveries from frontier AI labs, she says:
“Just one out of 1.3 million are actually exploitable.”
So why not fix everything? Because the business can’t stop. Her example is a bank where one application brings in 30% of revenue. Patching it all slows the very thing you are protecting.
The danger sits elsewhere. AI attackers do not work the way security teams do:
“They don’t think in silos. They think in actual chains.”
Defenders buy one tool for identity, another for phishing, another for vulnerabilities. An agent walks through all of them. A low-severity flaw you ignored becomes the link that holds the chain together.
That is her case for ending scan-and-patch. Find the one node that breaks the chain and fix that.
Her view of the AI version of Chernobyl is not comforting:
“When the Chernobyl will happen in cyber, it will be too late to fix it.”
The risk is not the critical bug on the dashboard. It is the boring one inside a chain.
This episode is presented by @juniper_square .
Full episode below 👇
youtu.be/FsmMLGSiP4A
See links for full episode:
🎧 Apple Podcasts: apple.co/4jtczCR
🟢 Spotify: bit.ly/3WVxZzR
1
68
Ethan Thornton (@ethanrthornton) on why ex-SpaceX engineers describe "needing to win" at their job.
His theory: that feeling only shows up when three things happen at once.
A mission tangible enough that your family understands why it matters. Real ownership, where the job is actually yours to run. And equity, because sovereignty in a capitalist system is capital.
Miss one of the three and you get competent employees. Get all three right and people chase the work.
His company went from $470M to $3.7B this year. He thinks that's the reason why.
The first Cold War was communism versus capitalism. This one is authoritarianism versus democracy, and it runs through capital, chips, oil, and social media all at once.
Ethan Thornton (@ethanrthornton) on why he's confident about one thing even if he won't predict when China moves on Taiwan:
"Democracy needs to beat totalitarianism."
1
4
540
John Borthwick (@Borthwick) on why "foundation model" is the right term: a foundation is something you build on top of, not something you live inside.
His bet on resilient AI isn't open vs closed. It's open and closed, local and cloud, general and highly verticalized, all layered together.
"In the last couple of months you could see it with the OpenAI agent hacks, the Hugging Face incidents, the WeChat issue two weeks ago. Each time, it got solved by open source and closed source models. We need both."
A stack with one point of failure isn't a stack. It's a single bet.
Ethan Thornton (@ethanrthornton) on what "founder mode" actually means:
"Founder mode is not jumping into the details and changing everything about your company. Founder mode is being super, super clear on setting the bar and finding people you trust."
His test for whether he's doing it right: a technician on his manufacturing floor has to feel comfortable walking up and telling him he's doing a shitty job as CEO.
"If I don't convince him, it means I'm probably doing the wrong thing."
Conviction without the ability to be told you're wrong isn't founder mode. It's just control.
John Borthwick (@Borthwick), on why AI won't replace human intelligence so much as expose how narrow our definition of it has become:
"When you think about the fact that you can have a baby, how do you describe that? We don't really have the words for that."
His point: the last fifty years of computing pushed us to define intelligence as efficiency. Language, logic, speed. The stuff a machine can now do better than us.
But that was always a reduction, not the full picture. The part of intelligence tied to physical experience, emotion, the things you feel but can't put into language, was never measured because nothing could compete with it.
"We've got this superhuman thing that can do a component of our intelligence, and we're gonna learn the other side of our intelligences."
AI isn't shrinking what it means to be intelligent. It's about to force us to finally define the rest of it.
John Borthwick's (@Borthwick) advice for white collar workers worried about AI: become a deep user, not a casual one.
Consumer apps like Instinct and Muse hand you agentic superpowers in a clean package. That's the point. Companies are built to package you a "thing" so you never look inside it.
His advice: use the packaged app first, then put it away. Try OpenClaw, try Hermes, build your own harness. You don't need to understand the neural network. You need to understand the layers of the package and what you can actually do with them.
Depth is becoming the new literacy.
5
384
Some services businesses can now hand 40 to 60% of their work to AI. THL’s Todd Abbrecht says buying them is really hard. One recent buy instead: a paving company.
Todd Abbrecht, co-CEO of THL Partners, on why the AI question in private equity is really a leverage question:
“Levered, long, and illiquid. On top of that, I can’t sell it. I can’t hedge it. I can’t get out of it. I have to be right on the buy.”
Take a white-collar services business where, in some cases he has seen, estimates put AI at 40 to 60% of the work. The owner might keep the savings. Or customers and competitors might take them. He says that is why valuing these businesses is so hard right now.
Venture can wear that uncertainty across a portfolio. A buyout owner can’t.
So he looks elsewhere. Healthcare. Pharma services. And, in his words:
“Places where you actually have physical hands.”
Which is how a firm with an AI playbook ends up owning a paving company:
“It is a place where AI isn’t going to disrupt that business.”
The edge in an AI market may not be predicting the winner. It may be owning something that doesn’t need predicting.
This episode is presented by @juniper_square.
Full episode below 👇
youtu.be/M6CuDJvuRAg
5
4
517
See links for full episode:
🎧 Apple Podcasts: tinyurl.com/2frvw5jw
🟢 Spotify: tinyurl.com/2u8m9d3r
1
91




