An inside look at technology, business, and culture with @davemorin @brit @lessin and @jessicalessin every Friday.
Joined June 2023
- Tweets551
- Following10
- Followers945
- Likes48
@lessin on the PHP vulnerability patched this week:
“Who’s using PHP? This is why, by the way, all of your regional banks shall be hacked.”
@brit: so where do we put our money, JP Morgan?
@davemorin: “In the ground.”
Then Dave on something he’s seen firsthand. He’s had stretches this year with access to frontier models that aren’t public.
“It’s really crazy, the differential between the feeling of using these things and what everybody’s talking about on the internet. There’s a wild gap. And I don’t even have access to the true frontier.”
And on the Roadster demo that landed days before Musk confirmed the merger committees: “That was literally somewhere two engineers going, hey, can we help justify merging these companies by sticking some rockets on a car?”
Dave: “But Sam, it’s genius.”
Sam: “It’s hilarious.”
New episode out soon.
People keep asking @lessin why he doesn’t restart Fin, the assistant company he ran with Andrew Kortina ten years ago. The models work now. Why not?
His answer is that nobody understands why they stopped. “It was not that we couldn’t provide a great service, and not that the economics didn’t work. Both of those were fine.”
Three problems, none of them technical.
You can’t just do scheduling. To be worth anything you do all the tasks, which makes you liable for every provider down the chain. You said a car would be there. It’s your problem now.
Mistakes cost more than your margin. DoorDash eats a botched food order because the margin’s fat and the food is cheap. Book the wrong $1,000 flight and you can’t absorb it.
And the tolerance is zero. “It’s so easy to fire a digital assistant. There’s no emotional bond. If your human assistant screws up, which they do all the time, you don’t instantly fire them.”
There’s always an error rate, 1% or 0.1%. “Can you make the economics of mistakes work?”
Today on the pod:
@Jessicalessin went full send on Muse and reports that about 70% of the use cases she was excited about have stopped working. Not because the model got worse. Good Eggs wants her to check out in the app, her kid's school portal refuses the login, and JP Morgan and Amazon are blocking agents outright.
@davemorin's version is worse. Instinct booked a dinner reservation in LA instead of San Francisco. He's out.
@lessin has seen this movie. He and Andrew Kortina ran Fin ten years ago and shut it down once they worked out that you become liable for every provider down the chain with no margin to cover your own mistakes. His call now is the death of the web, plus APIs that negotiate access back and forth like Hollywood agents.
Meanwhile Anthropic is meeting investors behind closed doors while they tape, the market's getting hammered, and @brit's three years of hip pain vanished in six days on peptides. Sam insists it's placebo.
Watch here on X or wherever you get your pods.
Topics:
01:09 Jess is back, Dave's birthday, and the October cold
03:04 Nasal rinses and fall immunity
04:02 Brit's peptide results: three years of pain gone in six days
06:43 Anthropic is meeting investors while we tape
07:34 The market is getting hammered: VIX, private credit, Apollo
08:16 Infrastructure IPOs stall and Nvidia becomes the last investor
10:01 Slowing revenue, CFOs cracking down, tokens moving to open models
10:36 Sam: would you rather be rich or right?
11:11 The models are a commodity and there is no lock-in
12:15 Brit won't give Muse her whole computer
13:41 Why Sam trusts Meta more than OpenAI or Anthropic
15:16 Jess: 70% of my Muse use cases don't work anymore
16:54 Dave: Instinct booked the restaurant in the wrong city
18:18 Sam on Fin, the assistant company he ran ten years ago
20:30 Why the economics of mistakes break agent services
22:31 Sam: what's inevitable is the death of the web
23:09 APIs as constant negotiation, and agents with reputations
24:51 Agent credit scores: FICO for bots
25:56 Amazon's Rufus still works, and other branding crimes
27:13 The seesaw: built up and torn down intraday
28:41 Gen Z media, Muse memes, and punk rock by another name
32:53 Is the punk rock thing just marketing?
34:37 Every company now needs a Gen Z meme lord
36:14 The truth subculture and who actually pays for it
37:32 Jess is interviewing Bari Weiss at WTF
40:20 Bumble goes private at $1.3B
43:07 Dave on seven years of Hinge and why dating is so hard
46:41 Apple's home hub and keeping screens away from kids
49:04 Everyone is going to see the Elizabeth Holmes movie
@brit asks whether the harnesses are going to have to cut individual deals with every major service provider.
@lessin thinks it goes further than deals. He thinks the API itself changes shape.
Right now an API gives you a fixed set of capabilities. His bet is that it becomes a back-and-forth.
Your agent shows up at Amazon and makes a case: I’m here for this customer, he’s great, he spends a lot, I’d like access to X, Y and Z.
Amazon counters by asking what it gets to know about him. They negotiate.
Which turns agent reputation into the asset. “Almost like Hollywood. If you have Ari Gold as your agent, you’re like, okay, this guy’s going to keep playing, he’s trustworthy, his clients are good.
Whereas if you show up as some open source agent I’ve never heard of, I’m not going to talk to you.”
The reason any of this happens: a perfect marketplace for the lowest price kills the travel industry. “No one’s going to survive that way.”
Today on the pod:
@Jessicalessin went full send on Muse and reports that about 70% of the use cases she was excited about have stopped working. Not because the model got worse. Good Eggs wants her to check out in the app, her kid's school portal refuses the login, and JP Morgan and Amazon are blocking agents outright.
@davemorin's version is worse. Instinct booked a dinner reservation in LA instead of San Francisco. He's out.
@lessin has seen this movie. He and Andrew Kortina ran Fin ten years ago and shut it down once they worked out that you become liable for every provider down the chain with no margin to cover your own mistakes. His call now is the death of the web, plus APIs that negotiate access back and forth like Hollywood agents.
Meanwhile Anthropic is meeting investors behind closed doors while they tape, the market's getting hammered, and @brit's three years of hip pain vanished in six days on peptides. Sam insists it's placebo.
Watch here on X or wherever you get your pods.
Topics:
01:09 Jess is back, Dave's birthday, and the October cold
03:04 Nasal rinses and fall immunity
04:02 Brit's peptide results: three years of pain gone in six days
06:43 Anthropic is meeting investors while we tape
07:34 The market is getting hammered: VIX, private credit, Apollo
08:16 Infrastructure IPOs stall and Nvidia becomes the last investor
10:01 Slowing revenue, CFOs cracking down, tokens moving to open models
10:36 Sam: would you rather be rich or right?
11:11 The models are a commodity and there is no lock-in
12:15 Brit won't give Muse her whole computer
13:41 Why Sam trusts Meta more than OpenAI or Anthropic
15:16 Jess: 70% of my Muse use cases don't work anymore
16:54 Dave: Instinct booked the restaurant in the wrong city
18:18 Sam on Fin, the assistant company he ran ten years ago
20:30 Why the economics of mistakes break agent services
22:31 Sam: what's inevitable is the death of the web
23:09 APIs as constant negotiation, and agents with reputations
24:51 Agent credit scores: FICO for bots
25:56 Amazon's Rufus still works, and other branding crimes
27:13 The seesaw: built up and torn down intraday
28:41 Gen Z media, Muse memes, and punk rock by another name
32:53 Is the punk rock thing just marketing?
34:37 Every company now needs a Gen Z meme lord
36:14 The truth subculture and who actually pays for it
37:32 Jess is interviewing Bari Weiss at WTF
40:20 Bumble goes private at $1.3B
43:07 Dave on seven years of Hinge and why dating is so hard
46:41 Apple's home hub and keeping screens away from kids
49:04 Everyone is going to see the Elizabeth Holmes movie
@brit says her WhatsApp groups are split clean down the middle on one question: how much are you giving your agents? No middle ground. People are either full send or absolutely not.
@lessin’s answer surprised me. Muse is the one he’s most willing to hand complete information to.
His reasoning comes from his father, Bob Lessin. Sam got scared on a roller coaster at Disney World as a kid and his dad told him not to worry, because Disney’s a public company.
“It’s a real company. They can be sued. It’s meaningful to them. They care about the security angles.”
Which is the thing he says the labs don’t have. “The problem with even OpenAI and Anthropic is if you sue them, it’s not clear you’ll ever get any money out of them. They might cause a national crisis, but you’re never going to sue them.”
Meta is legible, profitable and suable, with everything to lose. Same with Google.
He still doesn’t give any of them local access.
Today on the pod:
@Jessicalessin went full send on Muse and reports that about 70% of the use cases she was excited about have stopped working. Not because the model got worse. Good Eggs wants her to check out in the app, her kid's school portal refuses the login, and JP Morgan and Amazon are blocking agents outright.
@davemorin's version is worse. Instinct booked a dinner reservation in LA instead of San Francisco. He's out.
@lessin has seen this movie. He and Andrew Kortina ran Fin ten years ago and shut it down once they worked out that you become liable for every provider down the chain with no margin to cover your own mistakes. His call now is the death of the web, plus APIs that negotiate access back and forth like Hollywood agents.
Meanwhile Anthropic is meeting investors behind closed doors while they tape, the market's getting hammered, and @brit's three years of hip pain vanished in six days on peptides. Sam insists it's placebo.
Watch here on X or wherever you get your pods.
Topics:
01:09 Jess is back, Dave's birthday, and the October cold
03:04 Nasal rinses and fall immunity
04:02 Brit's peptide results: three years of pain gone in six days
06:43 Anthropic is meeting investors while we tape
07:34 The market is getting hammered: VIX, private credit, Apollo
08:16 Infrastructure IPOs stall and Nvidia becomes the last investor
10:01 Slowing revenue, CFOs cracking down, tokens moving to open models
10:36 Sam: would you rather be rich or right?
11:11 The models are a commodity and there is no lock-in
12:15 Brit won't give Muse her whole computer
13:41 Why Sam trusts Meta more than OpenAI or Anthropic
15:16 Jess: 70% of my Muse use cases don't work anymore
16:54 Dave: Instinct booked the restaurant in the wrong city
18:18 Sam on Fin, the assistant company he ran ten years ago
20:30 Why the economics of mistakes break agent services
22:31 Sam: what's inevitable is the death of the web
23:09 APIs as constant negotiation, and agents with reputations
24:51 Agent credit scores: FICO for bots
25:56 Amazon's Rufus still works, and other branding crimes
27:13 The seesaw: built up and torn down intraday
28:41 Gen Z media, Muse memes, and punk rock by another name
32:53 Is the punk rock thing just marketing?
34:37 Every company now needs a Gen Z meme lord
36:14 The truth subculture and who actually pays for it
37:32 Jess is interviewing Bari Weiss at WTF
40:20 Bumble goes private at $1.3B
43:07 Dave on seven years of Hinge and why dating is so hard
46:41 Apple's home hub and keeping screens away from kids
49:04 Everyone is going to see the Elizabeth Holmes movie
@Jessicalessin has been one of the biggest Muse users on this show. Not anymore, and not by choice.
“A solid 70% of my Muse use cases that I was so excited about do not work. I cannot complete things. The internet is just standing in my way 100% of the time.”
@lessin: “The internet’s fighting back.”
Her examples are mundane, which is what makes them bad news. Reordering from Good Eggs used to work every time. Now it tells her to enter payment details or check out in the app, three attempts running. Logging into her kids’ school portal, which she’s called her Grand AGI of Parenthood challenge, flat out refuses.
What still works: emailing the printer her kids’ piano practice sheets. No internet involved.
“I think it’s a great product and I’m still using it as much as I humanly can. But the magic of closing the loop in my core use cases is completely gone.”
@davemorin gives Meta real credit for how hard the build was, then names the trap. This category has to just work. Otherwise you lose people and churn people.
Today on the pod:
@Jessicalessin went full send on Muse and reports that about 70% of the use cases she was excited about have stopped working. Not because the model got worse. Good Eggs wants her to check out in the app, her kid's school portal refuses the login, and JP Morgan and Amazon are blocking agents outright.
@davemorin's version is worse. Instinct booked a dinner reservation in LA instead of San Francisco. He's out.
@lessin has seen this movie. He and Andrew Kortina ran Fin ten years ago and shut it down once they worked out that you become liable for every provider down the chain with no margin to cover your own mistakes. His call now is the death of the web, plus APIs that negotiate access back and forth like Hollywood agents.
Meanwhile Anthropic is meeting investors behind closed doors while they tape, the market's getting hammered, and @brit's three years of hip pain vanished in six days on peptides. Sam insists it's placebo.
Watch here on X or wherever you get your pods.
Topics:
01:09 Jess is back, Dave's birthday, and the October cold
03:04 Nasal rinses and fall immunity
04:02 Brit's peptide results: three years of pain gone in six days
06:43 Anthropic is meeting investors while we tape
07:34 The market is getting hammered: VIX, private credit, Apollo
08:16 Infrastructure IPOs stall and Nvidia becomes the last investor
10:01 Slowing revenue, CFOs cracking down, tokens moving to open models
10:36 Sam: would you rather be rich or right?
11:11 The models are a commodity and there is no lock-in
12:15 Brit won't give Muse her whole computer
13:41 Why Sam trusts Meta more than OpenAI or Anthropic
15:16 Jess: 70% of my Muse use cases don't work anymore
16:54 Dave: Instinct booked the restaurant in the wrong city
18:18 Sam on Fin, the assistant company he ran ten years ago
20:30 Why the economics of mistakes break agent services
22:31 Sam: what's inevitable is the death of the web
23:09 APIs as constant negotiation, and agents with reputations
24:51 Agent credit scores: FICO for bots
25:56 Amazon's Rufus still works, and other branding crimes
27:13 The seesaw: built up and torn down intraday
28:41 Gen Z media, Muse memes, and punk rock by another name
32:53 Is the punk rock thing just marketing?
34:37 Every company now needs a Gen Z meme lord
36:14 The truth subculture and who actually pays for it
37:32 Jess is interviewing Bari Weiss at WTF
40:20 Bumble goes private at $1.3B
43:07 Dave on seven years of Hinge and why dating is so hard
46:41 Apple's home hub and keeping screens away from kids
49:04 Everyone is going to see the Elizabeth Holmes movie
Today on the pod:
@Jessicalessin went full send on Muse and reports that about 70% of the use cases she was excited about have stopped working. Not because the model got worse. Good Eggs wants her to check out in the app, her kid's school portal refuses the login, and JP Morgan and Amazon are blocking agents outright.
@davemorin's version is worse. Instinct booked a dinner reservation in LA instead of San Francisco. He's out.
@lessin has seen this movie. He and Andrew Kortina ran Fin ten years ago and shut it down once they worked out that you become liable for every provider down the chain with no margin to cover your own mistakes. His call now is the death of the web, plus APIs that negotiate access back and forth like Hollywood agents.
Meanwhile Anthropic is meeting investors behind closed doors while they tape, the market's getting hammered, and @brit's three years of hip pain vanished in six days on peptides. Sam insists it's placebo.
Watch here on X or wherever you get your pods.
Topics:
01:09 Jess is back, Dave's birthday, and the October cold
03:04 Nasal rinses and fall immunity
04:02 Brit's peptide results: three years of pain gone in six days
06:43 Anthropic is meeting investors while we tape
07:34 The market is getting hammered: VIX, private credit, Apollo
08:16 Infrastructure IPOs stall and Nvidia becomes the last investor
10:01 Slowing revenue, CFOs cracking down, tokens moving to open models
10:36 Sam: would you rather be rich or right?
11:11 The models are a commodity and there is no lock-in
12:15 Brit won't give Muse her whole computer
13:41 Why Sam trusts Meta more than OpenAI or Anthropic
15:16 Jess: 70% of my Muse use cases don't work anymore
16:54 Dave: Instinct booked the restaurant in the wrong city
18:18 Sam on Fin, the assistant company he ran ten years ago
20:30 Why the economics of mistakes break agent services
22:31 Sam: what's inevitable is the death of the web
23:09 APIs as constant negotiation, and agents with reputations
24:51 Agent credit scores: FICO for bots
25:56 Amazon's Rufus still works, and other branding crimes
27:13 The seesaw: built up and torn down intraday
28:41 Gen Z media, Muse memes, and punk rock by another name
32:53 Is the punk rock thing just marketing?
34:37 Every company now needs a Gen Z meme lord
36:14 The truth subculture and who actually pays for it
37:32 Jess is interviewing Bari Weiss at WTF
40:20 Bumble goes private at $1.3B
43:07 Dave on seven years of Hinge and why dating is so hard
46:41 Apple's home hub and keeping screens away from kids
49:04 Everyone is going to see the Elizabeth Holmes movie
@Jessicalessin, on the day of taping: Anthropic is in closed-door meetings with investors right now about going public. The IPO’s still a go. The market’s getting hammered the same afternoon.
@lessin’s read on what’s underneath it. Meta turned Muse on as a consumer product and got real traction, their new models are competitive, and Gemini shipped everything at once.
“This stuff is commodity. The consumers are already at the point where it just doesn’t matter. The stuff is all good enough, which is classic race to the bottom. There is no lock in.”
Which breaks the pitch that funded all of it: “Anthropic’s and OpenAI’s whole pitch has been the infinity story of one company takes all, because they can’t possibly justify their existence if that’s not true.”
Then Jessica’s own experience, pointing the other way entirely. “A solid 70% of my Muse use cases that I was so excited about do not work.”
Sam: “I think it’s very hard for startups to survive what’s coming.”
New episode out soon.
@lessin tells every founder Slow invests in the same thing up front, and asks for it in writing: your seed pro rata will get challenged later.
“Some later stage funds are going to tell you that your seed pro rata doesn’t matter, and we’ll just quash it and not respect it, because they don’t need that. They already own plenty.”
It happened to Slow early on, with a large VC that was also one of their own LPs. @davemorin: “I remember this clear as day.”
The part worth sitting with is how routine it is. Seed funds get cut out of their pro rata constantly. Which is what makes a fund buying unexercised pro rata rights an odd thing to package, since plenty of people treat those rights as not worth the paper.
Meta and Microsoft have roughly halved how much their own employees use Claude, weeks out from the IPO. @davemorin’s read isn’t about price or benchmarks. Claude got over-tuned for parasocial relationships, people fell into something unhealthy with it, and organizations are now backing out.
@lessin runs Claude as core infrastructure and says switching off would be a one line change he’d never think about again.
Then it gets strange. A dinner where only the lab researchers raised their hands when asked whether one company would be left standing. A lab that believes the machine god will go back in time and simulate hell for anyone who didn’t help build it. And Sam’s case that the Valley caught zero-and-infinity thinking from crypto and broke its own spreadsheets.
@brit’s moderating. Jess is at an offsite, so it’s three VCs and no journalist.
Topics:
00:00 Trailer
00:51 Brit moderates: three VCs and no journalist
03:43 Peptide protocols, the Wolverine stack, and a Dexa scan
07:42 Sam on the Nathan Fielder movie
09:30 Meta and Microsoft halve their internal Claude usage
10:13 Dave: Claude was over-tuned for parasocial relationships
11:19 Sam: switching off Claude is a one line change
11:50 Silicon Valley is great at tech and bad at business
13:46 Zero, infinity, and what crypto taught the Valley
17:53 The dinner where only the lab people raised their hands
19:41 The lab that believes in the machine god
20:01 Lab leaks, infinite liability, and what a real framework looks like
21:56 The personal agent protocol: Meta, Walmart, Shopify, Stripe
23:01 Sam: nobody should care about MCP
24:03 Hilton traps Sam’s agent in a doom loop
26:30 If you ran Hilton, would you let the bots in?
27:15 Facebook Connect, Open Graph, and the missing value trade
29:41 Dave: building agents vs building agency
33:53 The social media era already ran this experiment
37:02 The Agent Treaty Framework is born
38:16 Why standards are over and treaties are next
41:28 Megalith and the market for unexercised pro rata
42:32 Sam’s pro rata war story
45:53 64% of LPs now want co-invest rights
48:03 What hedge fund letters say that VCs don’t
49:20 Main Street says it hates AI, then uses it
53:27 Plug of the week: the Mill
@lessin tried to book a hotel last minute. His agent got stuck in a loop on Hilton’s site and he gave up after three retries and booked it himself.
His read on why: “It’s Cloudflare. It’s basically businesses understanding correctly that if you get intermediated and become yet another way to clear a transaction at the back end, the market becomes too efficient, you’ll have no relationship with the customers, and your business is worthless.”
@brit points out Amazon did the same thing to Muse.
@davemorin thinks the new agent protocol exists to force this conversation. Sam doesn’t think a standard changes anything. Either you can be crawled or you can’t, and the entire textual internet already got taken: “Carmen Sandiego stole the entire internet to bootstrap these things.”
Brit asks whether Hilton even has the power to hold the line. Dave: “For sure they do.
Meta and Microsoft have roughly halved how much their own employees use Claude, weeks out from the IPO. @davemorin’s read isn’t about price or benchmarks. Claude got over-tuned for parasocial relationships, people fell into something unhealthy with it, and organizations are now backing out.
@lessin runs Claude as core infrastructure and says switching off would be a one line change he’d never think about again.
Then it gets strange. A dinner where only the lab researchers raised their hands when asked whether one company would be left standing. A lab that believes the machine god will go back in time and simulate hell for anyone who didn’t help build it. And Sam’s case that the Valley caught zero-and-infinity thinking from crypto and broke its own spreadsheets.
@brit’s moderating. Jess is at an offsite, so it’s three VCs and no journalist.
Topics:
00:00 Trailer
00:51 Brit moderates: three VCs and no journalist
03:43 Peptide protocols, the Wolverine stack, and a Dexa scan
07:42 Sam on the Nathan Fielder movie
09:30 Meta and Microsoft halve their internal Claude usage
10:13 Dave: Claude was over-tuned for parasocial relationships
11:19 Sam: switching off Claude is a one line change
11:50 Silicon Valley is great at tech and bad at business
13:46 Zero, infinity, and what crypto taught the Valley
17:53 The dinner where only the lab people raised their hands
19:41 The lab that believes in the machine god
20:01 Lab leaks, infinite liability, and what a real framework looks like
21:56 The personal agent protocol: Meta, Walmart, Shopify, Stripe
23:01 Sam: nobody should care about MCP
24:03 Hilton traps Sam’s agent in a doom loop
26:30 If you ran Hilton, would you let the bots in?
27:15 Facebook Connect, Open Graph, and the missing value trade
29:41 Dave: building agents vs building agency
33:53 The social media era already ran this experiment
37:02 The Agent Treaty Framework is born
38:16 Why standards are over and treaties are next
41:28 Megalith and the market for unexercised pro rata
42:32 Sam’s pro rata war story
45:53 64% of LPs now want co-invest rights
48:03 What hedge fund letters say that VCs don’t
49:20 Main Street says it hates AI, then uses it
53:27 Plug of the week: the Mill
@davemorin: “One of the labs, I’ll let everyone guess which one, believes that if you don’t bring about the machine God, then the machine God, once it is created, will go back in time and put you into a simulation of hell. This is a real belief.”
@lessin’s response is that this is the whole problem. A religious zero-and-infinity infestation that’s producing genuinely bad decisions.
His case for what we should be looking at instead: lab leaks, which have already happened once at Hugging Face. Trillions in damage, Katrina-scale. Banks going down, systems frozen, people dying.
And unlike the machine God, it’s a problem with an obvious shape. Post-2008 banks had to hold cash reserves. If you’re running a lab and something gets out but your balance sheet can’t cover it, you shouldn’t be dumping that on the public.
“Because we’re so obsessed with the tails of zero and infinity and the religion, we’re missing actually good decisions about how to operate in a new technological landscape.”
Meta and Microsoft have roughly halved how much their own employees use Claude, weeks out from the IPO. @davemorin’s read isn’t about price or benchmarks. Claude got over-tuned for parasocial relationships, people fell into something unhealthy with it, and organizations are now backing out.
@lessin runs Claude as core infrastructure and says switching off would be a one line change he’d never think about again.
Then it gets strange. A dinner where only the lab researchers raised their hands when asked whether one company would be left standing. A lab that believes the machine god will go back in time and simulate hell for anyone who didn’t help build it. And Sam’s case that the Valley caught zero-and-infinity thinking from crypto and broke its own spreadsheets.
@brit’s moderating. Jess is at an offsite, so it’s three VCs and no journalist.
Topics:
00:00 Trailer
00:51 Brit moderates: three VCs and no journalist
03:43 Peptide protocols, the Wolverine stack, and a Dexa scan
07:42 Sam on the Nathan Fielder movie
09:30 Meta and Microsoft halve their internal Claude usage
10:13 Dave: Claude was over-tuned for parasocial relationships
11:19 Sam: switching off Claude is a one line change
11:50 Silicon Valley is great at tech and bad at business
13:46 Zero, infinity, and what crypto taught the Valley
17:53 The dinner where only the lab people raised their hands
19:41 The lab that believes in the machine god
20:01 Lab leaks, infinite liability, and what a real framework looks like
21:56 The personal agent protocol: Meta, Walmart, Shopify, Stripe
23:01 Sam: nobody should care about MCP
24:03 Hilton traps Sam’s agent in a doom loop
26:30 If you ran Hilton, would you let the bots in?
27:15 Facebook Connect, Open Graph, and the missing value trade
29:41 Dave: building agents vs building agency
33:53 The social media era already ran this experiment
37:02 The Agent Treaty Framework is born
38:16 Why standards are over and treaties are next
41:28 Megalith and the market for unexercised pro rata
42:32 Sam’s pro rata war story
45:53 64% of LPs now want co-invest rights
48:03 What hedge fund letters say that VCs don’t
49:20 Main Street says it hates AI, then uses it
53:27 Plug of the week: the Mill
Meta and Microsoft have roughly halved how much their own staff use Claude, and both are pushing people onto in-house models instead.
@lessin: “The end state of all things is narcissism, where you just talk to yourself. Muse talks to Muse.”
@davemorin’s diagnosis isn’t cost or benchmarks. “There’s a certain flavor to Claude slop that you can actually sense. You can tell who on Twitter is talking to Claude too much.”
He thinks the real damage is upstream of all the doom talk: “It was over tuned for parasocial relationship. People got deeply into relationship with it in a way that was pretty unhealthy. And I think people are waking up to it now, both individually and collectively as organizations.”
Sam uses a lot of Claude and doesn’t talk to it that way at all. For him it’s infrastructure, it’s excellent, and that’s the problem. “It’s very hard to build important businesses for the long term when if it went away tomorrow, there’s a perfect replacement and always will be.”
He isn’t switching. He also says it’d be a one line change if he did.
Meta and Microsoft have roughly halved how much their own employees use Claude, weeks out from the IPO. @davemorin’s read isn’t about price or benchmarks. Claude got over-tuned for parasocial relationships, people fell into something unhealthy with it, and organizations are now backing out.
@lessin runs Claude as core infrastructure and says switching off would be a one line change he’d never think about again.
Then it gets strange. A dinner where only the lab researchers raised their hands when asked whether one company would be left standing. A lab that believes the machine god will go back in time and simulate hell for anyone who didn’t help build it. And Sam’s case that the Valley caught zero-and-infinity thinking from crypto and broke its own spreadsheets.
@brit’s moderating. Jess is at an offsite, so it’s three VCs and no journalist.
Topics:
00:00 Trailer
00:51 Brit moderates: three VCs and no journalist
03:43 Peptide protocols, the Wolverine stack, and a Dexa scan
07:42 Sam on the Nathan Fielder movie
09:30 Meta and Microsoft halve their internal Claude usage
10:13 Dave: Claude was over-tuned for parasocial relationships
11:19 Sam: switching off Claude is a one line change
11:50 Silicon Valley is great at tech and bad at business
13:46 Zero, infinity, and what crypto taught the Valley
17:53 The dinner where only the lab people raised their hands
19:41 The lab that believes in the machine god
20:01 Lab leaks, infinite liability, and what a real framework looks like
21:56 The personal agent protocol: Meta, Walmart, Shopify, Stripe
23:01 Sam: nobody should care about MCP
24:03 Hilton traps Sam’s agent in a doom loop
26:30 If you ran Hilton, would you let the bots in?
27:15 Facebook Connect, Open Graph, and the missing value trade
29:41 Dave: building agents vs building agency
33:53 The social media era already ran this experiment
37:02 The Agent Treaty Framework is born
38:16 Why standards are over and treaties are next
41:28 Megalith and the market for unexercised pro rata
42:32 Sam’s pro rata war story
45:53 64% of LPs now want co-invest rights
48:03 What hedge fund letters say that VCs don’t
49:20 Main Street says it hates AI, then uses it
53:27 Plug of the week: the Mill
Silicon Valley started putting infinity into its Excel models.
That’s @lessin’s diagnosis on this week’s episode with @davemorin and @brit. It came up while the hosts were arguing that the Valley is great at technology and bad at business.
His argument: the West has never been comfortable with zero or infinity. Zero came from India, and the church fought it until it was too useful to keep out.
Normal business math runs on competition and compounding. Put a zero or an infinity anywhere in the chain and the tail dominates the whole story.
Google and Facebook taught the Valley that companies could be unimaginably big. Bitcoin taught it that belief alone could make something worth everything or nothing.
Sam’s read: that’s the lens for understanding the AI labs and the capital chasing them.
Meta and Microsoft have roughly halved how much their own employees use Claude, weeks out from the IPO. @davemorin’s read isn’t about price or benchmarks. Claude got over-tuned for parasocial relationships, people fell into something unhealthy with it, and organizations are now backing out.
@lessin runs Claude as core infrastructure and says switching off would be a one line change he’d never think about again.
Then it gets strange. A dinner where only the lab researchers raised their hands when asked whether one company would be left standing. A lab that believes the machine god will go back in time and simulate hell for anyone who didn’t help build it. And Sam’s case that the Valley caught zero-and-infinity thinking from crypto and broke its own spreadsheets.
@brit’s moderating. Jess is at an offsite, so it’s three VCs and no journalist.
Topics:
00:00 Trailer
00:51 Brit moderates: three VCs and no journalist
03:43 Peptide protocols, the Wolverine stack, and a Dexa scan
07:42 Sam on the Nathan Fielder movie
09:30 Meta and Microsoft halve their internal Claude usage
10:13 Dave: Claude was over-tuned for parasocial relationships
11:19 Sam: switching off Claude is a one line change
11:50 Silicon Valley is great at tech and bad at business
13:46 Zero, infinity, and what crypto taught the Valley
17:53 The dinner where only the lab people raised their hands
19:41 The lab that believes in the machine god
20:01 Lab leaks, infinite liability, and what a real framework looks like
21:56 The personal agent protocol: Meta, Walmart, Shopify, Stripe
23:01 Sam: nobody should care about MCP
24:03 Hilton traps Sam’s agent in a doom loop
26:30 If you ran Hilton, would you let the bots in?
27:15 Facebook Connect, Open Graph, and the missing value trade
29:41 Dave: building agents vs building agency
33:53 The social media era already ran this experiment
37:02 The Agent Treaty Framework is born
38:16 Why standards are over and treaties are next
41:28 Megalith and the market for unexercised pro rata
42:32 Sam’s pro rata war story
45:53 64% of LPs now want co-invest rights
48:03 What hedge fund letters say that VCs don’t
49:20 Main Street says it hates AI, then uses it
53:27 Plug of the week: the Mill
@brit’s moderating this week. Jessica’s at an offsite, so it’s three VCs and no journalist.
She opens with the number: Meta and Microsoft, two of Anthropic’s largest enterprise clients, have roughly halved how much their own staff use Claude and pushed them onto in-house models.
@lessin: “Claude must be so lonely.”
Then @davemorin raises something stranger. One of the labs, and he won’t say which, holds the belief that if you don’t bring about the machine God, then once it exists it goes back in time and puts you in a simulation of hell.
“This is a real belief.”
New episode live soon.
@adrianaoun grew up with a thick accent and didn't feel like he fit in, so he went online, where he could be as tall as he wanted and have no accent at all.
That's his case for why agent-to-agent interaction stops feeling dystopian. We're at a local optimum. Two things shift it.
One, consumers only adopt what they want, so "products have to be better and better at kind of tricking you into believing this is what you want."
Two, we get used to things. "If I told you to imagine the year 2026 in the year 1950, we're all going to be walking around staring at four inch pieces of glass, you'd be like, that's insane, nobody would ever do that. Except that's literally what we do all day and we feel empowered by doing it."
He's making the optimistic argument. The word he reaches for is tricking.
Anthropic's S-1 leaked and the numbers don't match the narrative. Revenue flatlining as token prices fall, SaaS investors posting memes welcoming them to the no-growth table. @Jessicalessin still puts it at 60% they get out in November, which looks brave in a week when Oura pulled its own IPO and blamed market conditions.
@lessin's problem with the core business is simpler than any of that. Claude is the center of his infrastructure. "If I woke up tomorrow and Claude was gone, I would care for about one second."
Full house plus @adrianaoun, who drops in to argue there's no moat left unless you own the infrastructure, the data, or the algorithms.
Also: @brit's agent started coordinating date night with @davemorin's agent, and she stopped texting her husband for a day.
Recorded live from the X boardroom. Watch here or wherever you get your pods.
Topics:
00:00 Trailer
01:04 Live from X HQ, and a 39 degree cold plunge
02:57 Dunbar's number, but for bots
05:48 Anthropic's leaked S-1 and the flatlining revenue
07:37 Oura pulls its IPO and the LPs get stuck
09:36 What's actually happening in the markets
10:41 Wall Street and Silicon Valley split on AI's price tag
12:45 World Labs sells for $8B: the barbell strategy
14:11 Half a trillion in contracts, and who's really making money
14:54 What happens if OpenAI and Anthropic don't work out?
16:11 Priced for a growth rate bigger than the US economy
18:04 Open models pass closed models on the charts
18:51 Sam: if Claude vanished tomorrow, I'd care for one second
19:43 The forward deployed bet and the luxury token problem
22:16 Adrian drops in: 300 startups, 40 unicorns, and a plane
24:39 Jess goes full Muse
26:28 The bot AGI challenge: after-school plans and the mom index
28:23 OpenAI rebrands Codex as Dots
30:30 Adrian on the moat: winner take some
31:38 Brit's agent starts talking to Dave's agent
33:51 No switching costs, so BD deals are the only lock-in
36:13 Dave: Instinct is the new Kayak
36:57 Maybe it's the beginning of times, not the end
38:47 Adrian's opposite thesis: we go further into the screens
43:39 Who has agency, you or the technology?
44:22 Offline is the new luxury
46:31 Everyone is building personal apps now
47:54 Book rec: Ari Emanuel's Roll the Calls
50:44 Adrian's parting shot: infrastructure, data, or algorithms
53:02 Sam's Middle Ages 2.0: castles, moats, and hoarding
@davemorin's case for optimism starts as a criticism of the last fifty years.
Since the late seventies people have sat in buildings staring at glowing rectangles, fidgeting with computers to make them do what they wanted.
"Most of the output has been not that great. It didn't really produce that much interesting cultural content. It didn't really produce the greatest architecture."
He thinks that era's ending now that you can just speak English to these things.
Then he brings up the film people name when they describe the future they want, which is Her.
@lessin: "That is like a very cynical movie. The whole point is that guy's so lonely."
Dave says nobody citing it means the ending. They mean the apartment, which has no screens in it anywhere.
Anthropic's S-1 leaked and the numbers don't match the narrative. Revenue flatlining as token prices fall, SaaS investors posting memes welcoming them to the no-growth table. @Jessicalessin still puts it at 60% they get out in November, which looks brave in a week when Oura pulled its own IPO and blamed market conditions.
@lessin's problem with the core business is simpler than any of that. Claude is the center of his infrastructure. "If I woke up tomorrow and Claude was gone, I would care for about one second."
Full house plus @adrianaoun, who drops in to argue there's no moat left unless you own the infrastructure, the data, or the algorithms.
Also: @brit's agent started coordinating date night with @davemorin's agent, and she stopped texting her husband for a day.
Recorded live from the X boardroom. Watch here or wherever you get your pods.
Topics:
00:00 Trailer
01:04 Live from X HQ, and a 39 degree cold plunge
02:57 Dunbar's number, but for bots
05:48 Anthropic's leaked S-1 and the flatlining revenue
07:37 Oura pulls its IPO and the LPs get stuck
09:36 What's actually happening in the markets
10:41 Wall Street and Silicon Valley split on AI's price tag
12:45 World Labs sells for $8B: the barbell strategy
14:11 Half a trillion in contracts, and who's really making money
14:54 What happens if OpenAI and Anthropic don't work out?
16:11 Priced for a growth rate bigger than the US economy
18:04 Open models pass closed models on the charts
18:51 Sam: if Claude vanished tomorrow, I'd care for one second
19:43 The forward deployed bet and the luxury token problem
22:16 Adrian drops in: 300 startups, 40 unicorns, and a plane
24:39 Jess goes full Muse
26:28 The bot AGI challenge: after-school plans and the mom index
28:23 OpenAI rebrands Codex as Dots
30:30 Adrian on the moat: winner take some
31:38 Brit's agent starts talking to Dave's agent
33:51 No switching costs, so BD deals are the only lock-in
36:13 Dave: Instinct is the new Kayak
36:57 Maybe it's the beginning of times, not the end
38:47 Adrian's opposite thesis: we go further into the screens
43:39 Who has agency, you or the technology?
44:22 Offline is the new luxury
46:31 Everyone is building personal apps now
47:54 Book rec: Ari Emanuel's Roll the Calls
50:44 Adrian's parting shot: infrastructure, data, or algorithms
53:02 Sam's Middle Ages 2.0: castles, moats, and hoarding
@brit connected her agent to @davemorin's so the two could sort out date night. It worked. A day later she realized she hadn't texted her husband.
"I did it with Dave and then I didn't even text with Dave for a whole day. I was just like, Dave, text my agent."
She isn't against the thing. She's in a set of women's WhatsApp groups, mostly mothers and founders, where everyone's obsessed with Instinct for exactly this feature, because the coordination it absorbs is annoying work nobody wants.
@adrianaoun doesn't buy the worry. "Some of my best friends in the world I haven't seen for like nine months, but I talk all the time through FaceTime.
Why is FaceTime okay but my instinct agent is not okay?" He thinks the Overton window hasn't moved yet.
Brit: "There's no humanity in it. I'm losing actual context of humans."
Anthropic's S-1 leaked and the numbers don't match the narrative. Revenue flatlining as token prices fall, SaaS investors posting memes welcoming them to the no-growth table. @Jessicalessin still puts it at 60% they get out in November, which looks brave in a week when Oura pulled its own IPO and blamed market conditions.
@lessin's problem with the core business is simpler than any of that. Claude is the center of his infrastructure. "If I woke up tomorrow and Claude was gone, I would care for about one second."
Full house plus @adrianaoun, who drops in to argue there's no moat left unless you own the infrastructure, the data, or the algorithms.
Also: @brit's agent started coordinating date night with @davemorin's agent, and she stopped texting her husband for a day.
Recorded live from the X boardroom. Watch here or wherever you get your pods.
Topics:
00:00 Trailer
01:04 Live from X HQ, and a 39 degree cold plunge
02:57 Dunbar's number, but for bots
05:48 Anthropic's leaked S-1 and the flatlining revenue
07:37 Oura pulls its IPO and the LPs get stuck
09:36 What's actually happening in the markets
10:41 Wall Street and Silicon Valley split on AI's price tag
12:45 World Labs sells for $8B: the barbell strategy
14:11 Half a trillion in contracts, and who's really making money
14:54 What happens if OpenAI and Anthropic don't work out?
16:11 Priced for a growth rate bigger than the US economy
18:04 Open models pass closed models on the charts
18:51 Sam: if Claude vanished tomorrow, I'd care for one second
19:43 The forward deployed bet and the luxury token problem
22:16 Adrian drops in: 300 startups, 40 unicorns, and a plane
24:39 Jess goes full Muse
26:28 The bot AGI challenge: after-school plans and the mom index
28:23 OpenAI rebrands Codex as Dots
30:30 Adrian on the moat: winner take some
31:38 Brit's agent starts talking to Dave's agent
33:51 No switching costs, so BD deals are the only lock-in
36:13 Dave: Instinct is the new Kayak
36:57 Maybe it's the beginning of times, not the end
38:47 Adrian's opposite thesis: we go further into the screens
43:39 Who has agency, you or the technology?
44:22 Offline is the new luxury
46:31 Everyone is building personal apps now
47:54 Book rec: Ari Emanuel's Roll the Calls
50:44 Adrian's parting shot: infrastructure, data, or algorithms
53:02 Sam's Middle Ages 2.0: castles, moats, and hoarding
@lessin on why everyone is quietly rooting for AI: "It's in no one's incentive for these things to not work out, even if you're not invested in them. And yet there's a real risk they're not going to work out. That's a really weird place to be."
His frame for it is the triangle trade.
DC needs it, because the country needs 8% GDP growth it doesn't have and has a debt problem it can't grow out of otherwise. Silicon Valley is having a party.
And New York needs it in a way it never needed Facebook or Google, because those companies didn't require much capital.
This one requires everything. Every insurance product, every financing structure.
"If you did not have an AI story, we would be in terribly dire straits."
Anthropic's S-1 leaked and the numbers don't match the narrative. Revenue flatlining as token prices fall, SaaS investors posting memes welcoming them to the no-growth table. @Jessicalessin still puts it at 60% they get out in November, which looks brave in a week when Oura pulled its own IPO and blamed market conditions.
@lessin's problem with the core business is simpler than any of that. Claude is the center of his infrastructure. "If I woke up tomorrow and Claude was gone, I would care for about one second."
Full house plus @adrianaoun, who drops in to argue there's no moat left unless you own the infrastructure, the data, or the algorithms.
Also: @brit's agent started coordinating date night with @davemorin's agent, and she stopped texting her husband for a day.
Recorded live from the X boardroom. Watch here or wherever you get your pods.
Topics:
00:00 Trailer
01:04 Live from X HQ, and a 39 degree cold plunge
02:57 Dunbar's number, but for bots
05:48 Anthropic's leaked S-1 and the flatlining revenue
07:37 Oura pulls its IPO and the LPs get stuck
09:36 What's actually happening in the markets
10:41 Wall Street and Silicon Valley split on AI's price tag
12:45 World Labs sells for $8B: the barbell strategy
14:11 Half a trillion in contracts, and who's really making money
14:54 What happens if OpenAI and Anthropic don't work out?
16:11 Priced for a growth rate bigger than the US economy
18:04 Open models pass closed models on the charts
18:51 Sam: if Claude vanished tomorrow, I'd care for one second
19:43 The forward deployed bet and the luxury token problem
22:16 Adrian drops in: 300 startups, 40 unicorns, and a plane
24:39 Jess goes full Muse
26:28 The bot AGI challenge: after-school plans and the mom index
28:23 OpenAI rebrands Codex as Dots
30:30 Adrian on the moat: winner take some
31:38 Brit's agent starts talking to Dave's agent
33:51 No switching costs, so BD deals are the only lock-in
36:13 Dave: Instinct is the new Kayak
36:57 Maybe it's the beginning of times, not the end
38:47 Adrian's opposite thesis: we go further into the screens
43:39 Who has agency, you or the technology?
44:22 Offline is the new luxury
46:31 Everyone is building personal apps now
47:54 Book rec: Ari Emanuel's Roll the Calls
50:44 Adrian's parting shot: infrastructure, data, or algorithms
53:02 Sam's Middle Ages 2.0: castles, moats, and hoarding
@davemorin asks the question everyone skips: what actually happens if these companies don't work out? Do they just shut down?
@Jessicalessin's answer is that the first shoe to drop is pricing, not survival. If public markets value you at a fraction of what you last raised at privately, that's a real event. For employees especially.
It changes the narrative."Many high flying companies have come out at one valuation and settled somewhere else. And I think that probably is the most likely thing."
Early investors take their bag, the thing reprices, and liability, regulation and growth are all jump balls from there. Not a death sentence. Also not comfortable.
Her bottom line: AI is important enough and these opportunities are large enough to build real businesses.
"But are they valued at a price that assumes a stratospheric growth rate across all industries for all time? Probably not."
Anthropic's S-1 leaked and the numbers don't match the narrative. Revenue flatlining as token prices fall, SaaS investors posting memes welcoming them to the no-growth table. @Jessicalessin still puts it at 60% they get out in November, which looks brave in a week when Oura pulled its own IPO and blamed market conditions.
@lessin's problem with the core business is simpler than any of that. Claude is the center of his infrastructure. "If I woke up tomorrow and Claude was gone, I would care for about one second."
Full house plus @adrianaoun, who drops in to argue there's no moat left unless you own the infrastructure, the data, or the algorithms.
Also: @brit's agent started coordinating date night with @davemorin's agent, and she stopped texting her husband for a day.
Recorded live from the X boardroom. Watch here or wherever you get your pods.
Topics:
00:00 Trailer
01:04 Live from X HQ, and a 39 degree cold plunge
02:57 Dunbar's number, but for bots
05:48 Anthropic's leaked S-1 and the flatlining revenue
07:37 Oura pulls its IPO and the LPs get stuck
09:36 What's actually happening in the markets
10:41 Wall Street and Silicon Valley split on AI's price tag
12:45 World Labs sells for $8B: the barbell strategy
14:11 Half a trillion in contracts, and who's really making money
14:54 What happens if OpenAI and Anthropic don't work out?
16:11 Priced for a growth rate bigger than the US economy
18:04 Open models pass closed models on the charts
18:51 Sam: if Claude vanished tomorrow, I'd care for one second
19:43 The forward deployed bet and the luxury token problem
22:16 Adrian drops in: 300 startups, 40 unicorns, and a plane
24:39 Jess goes full Muse
26:28 The bot AGI challenge: after-school plans and the mom index
28:23 OpenAI rebrands Codex as Dots
30:30 Adrian on the moat: winner take some
31:38 Brit's agent starts talking to Dave's agent
33:51 No switching costs, so BD deals are the only lock-in
36:13 Dave: Instinct is the new Kayak
36:57 Maybe it's the beginning of times, not the end
38:47 Adrian's opposite thesis: we go further into the screens
43:39 Who has agency, you or the technology?
44:22 Offline is the new luxury
46:31 Everyone is building personal apps now
47:54 Book rec: Ari Emanuel's Roll the Calls
50:44 Adrian's parting shot: infrastructure, data, or algorithms
53:02 Sam's Middle Ages 2.0: castles, moats, and hoarding
@lessin uses Claude as the core of his infrastructure. He also says this:
"If I woke up tomorrow and Claude was gone, I would care for about one second."
That's the whole commoditization argument in one sentence, from a heavy user. His read on the core business: can you sell premium tokens at a higher price than the open models, on the theory that what you're doing is so valuable people will sink infinite money into marginally better tokens?
"That's been totally true for the last number of months. It's flatlining."
And Meta's Muse, which he'll grant is less advanced than Fable, is clearly sufficient for what normal people actually need. Which puts a dent in the story.
Anthropic's S-1 leaked and the numbers don't match the narrative. Revenue flatlining as token prices fall, SaaS investors posting memes welcoming them to the no-growth table. @Jessicalessin still puts it at 60% they get out in November, which looks brave in a week when Oura pulled its own IPO and blamed market conditions.
@lessin's problem with the core business is simpler than any of that. Claude is the center of his infrastructure. "If I woke up tomorrow and Claude was gone, I would care for about one second."
Full house plus @adrianaoun, who drops in to argue there's no moat left unless you own the infrastructure, the data, or the algorithms.
Also: @brit's agent started coordinating date night with @davemorin's agent, and she stopped texting her husband for a day.
Recorded live from the X boardroom. Watch here or wherever you get your pods.
Topics:
00:00 Trailer
01:04 Live from X HQ, and a 39 degree cold plunge
02:57 Dunbar's number, but for bots
05:48 Anthropic's leaked S-1 and the flatlining revenue
07:37 Oura pulls its IPO and the LPs get stuck
09:36 What's actually happening in the markets
10:41 Wall Street and Silicon Valley split on AI's price tag
12:45 World Labs sells for $8B: the barbell strategy
14:11 Half a trillion in contracts, and who's really making money
14:54 What happens if OpenAI and Anthropic don't work out?
16:11 Priced for a growth rate bigger than the US economy
18:04 Open models pass closed models on the charts
18:51 Sam: if Claude vanished tomorrow, I'd care for one second
19:43 The forward deployed bet and the luxury token problem
22:16 Adrian drops in: 300 startups, 40 unicorns, and a plane
24:39 Jess goes full Muse
26:28 The bot AGI challenge: after-school plans and the mom index
28:23 OpenAI rebrands Codex as Dots
30:30 Adrian on the moat: winner take some
31:38 Brit's agent starts talking to Dave's agent
33:51 No switching costs, so BD deals are the only lock-in
36:13 Dave: Instinct is the new Kayak
36:57 Maybe it's the beginning of times, not the end
38:47 Adrian's opposite thesis: we go further into the screens
43:39 Who has agency, you or the technology?
44:22 Offline is the new luxury
46:31 Everyone is building personal apps now
47:54 Book rec: Ari Emanuel's Roll the Calls
50:44 Adrian's parting shot: infrastructure, data, or algorithms
53:02 Sam's Middle Ages 2.0: castles, moats, and hoarding