There is no moat in AI agents. None. I'll die on this hill — new More or Less is up. I switch between Muse, Instinct, and Dots at basically zero cost: one GitHub repo, a few databases, feed the new thing my context, done. If the cost of writing software is going to zero, API integrations are not a moat. The only moat left is user setup cost — which means winner-take-some, not winner-take-all. Adrian's point: AI collapses value the second an idea works. Jev got reverse-engineered in 48 hours and the people who built it capture nothing. My take: play that to the limit and you get Middle Ages 2.0 — everyone builds a castle, hoards data, ships black boxes that self-destruct if you open them. Dave says humans don't work that way. We'll see. Meanwhile Brit's Instinct started talking to Dave's Instinct and they didn't text for a day. She says it's killing humanity. Nobody said that about FaceTime. And Jess announced the only benchmark that matters: log into school software, sync three kids' after-school schedules. Muse: 90%. Every other eval is bullshit.

This week's More or Less got heated. Jess said she'd rather send our kids to the military than to a16z's new university. Brit's take: the $35M 'school' is a marketing spend — top-of-funnel to sign founders before they're 19. I'm all for burning down broken higher ed, but an advisory page of Altman and Murati is not a substitute for education. Fight me. Also: a Blackstone exec at The Information's AI summit got asked point blank who pays back all the data center debt. Answer: 'We don't know.' Not embarrassed. Just where we are in the cycle. Jess nailed the deeper problem — the AGI trade is the St. Petersburg paradox. If you sell an infinite payout, there is no price you can't justify. Now the story is quietly becoming 'AGI is gradual'... conveniently after AGI clauses got written into billion-dollar agreements because deals were moving too fast to do anything else. Infinite demand ≠ infinite margins. Calculations aren't valuable just because you consume more of them. Oh — and Mazzeo claims GLP-1 revenue now rivals AI revenue. Ep 168 is up.

@lessin's theory of the era, delivered while four people compare sleep scores on a podcast: "If you want to understand our era, looking back on our years, we are entertaining ourselves to death. This is the most entertaining business cycle in the history of the world." His point is that AI turns out to be extremely good at entertainment, so everything now organizes around producing the most entertaining version of events. Including the AI doom discourse itself. @davemorin's line is the sharpest one in the segment: longevity age scores are the doom tourism of health apps. Then Sam, on the app scores: "Aura is kind of reasonable because they're selling you some semblance of reality. And Eight Sleep is the most self-congratulatory." @brit : "Okay, now Eight Sleep's dumb. I don't believe any of this." Dave: "Once Brit's losing, she just stops playing the game."

SF VC'S have found a new watering hole in Jackson Square. 37+ firms in about six blocks. Fun fact: These are 170 year old gold rush brick buildings and @ThriveCapital paid $9M for a firehouse at 451 Pacific.

This isn't even a nike commercial and it makes me want to buy nike stock. pure competition against yourself, pure achievement for the sake of it. Confidence on the start, walk to the start line saying i am going to win... keep it rolling.

Pearl (@prlnet) has the fastest FP8 MoE Kernels for Blackwell chips! Group-Matrix-Multiplication is one of the most optimized operations in AI. As a former researcher at @nvidia I can say firsthand that pushing the performance frontier of MatMuls is... Hard. Pearl-GEMM is ~5% faster than prior state-of-art Quack (@tri_dao) and ~42% faster than Flashinfer on B200s.