More or Less · Topics · Episode #169

If Claude Vanished Tomorrow, Sam Would Care for "About One Second"

Clip from the full recording, 20:16–24:43 · download mp4 · watch at this point on YouTube

Open models are outpacing closed ones, Dave reveals the forward-deployed token trap, and Sam says the core Anthropic business has a baseline problem: users barely care which model they use.

Brit opens with the Vercel charts showing open models gaining on closed ones, and Dave counters that the pie is also getting much bigger, so the chart may not matter to public-market outcomes. Sam breaks any large public AI company into a core business plus option value, and the core question for Anthropic is whether it can sell premium tokens at a price above open models. That story was true for months, he says, and now it is flatlining, and models like Muse are clearly sufficient for what normal people need.

Dave adds a nuance he hasn’t seen anyone discuss: the labs built forward-deployed teams to act like McKinsey for the Fortune 2000, expecting customers to burn luxury tokens. Then open models came out of nowhere and CFOs started asking why they should spend the expensive tokens at all. Sam says forward deployment makes sense as lock-in in the Palantir model, but it is a very different business if you only get paid on tokens.

Sam then gives his biggest takeaway: he uses Claude constantly and it is the core of his infrastructure brain, but if it vanished tomorrow he would care for about one second. There are option values, but you have to support a massive core cost structure, and the question is whether there is a core business underneath.

Key points

  • Brit: Vercel charts show open models outpacing closed ones, another headwind for the labs.
  • Dave: the pie is also much bigger, so the open-versus-closed chart can be read two ways.
  • Sam: a public AI company is a core business plus option value; the core is selling premium tokens above open-model prices, and it is flatlining.
  • Dave: forward-deployed teams were meant to lock in Fortune 500 token spend, but CFOs now want the cheaper open tokens.
  • Sam: forward deployment can create lock-in as in Palantir, but it's a different model when you get paid on tokens.
  • Sam: if Claude disappeared tomorrow he'd care for about a second; that's the baseline problem in the core business.

Where they landed

BritOpen models gaining on closed ones is a headwind; Meta just hiring a head of enterprise adds competition.
DaveThe forward-deployed strategy may have backfired as CFOs defect to cheap open tokens.
SamCore business is premium tokens, and users would barely notice if Claude vanished; option value doesn't pay for the cost structure.
JessicaAgrees Sam would be fine without Claude.

Quotes

“the displacement of closed versus open continuing to be a headwind as well.”— Brit
“the open models came out of nowhere. And all of a sudden the CFOs were like, hey, you built me this software to spend your tokens, but like your tokens are really expensive.”— Dave
“if I woke up tomorrow and Claude was gone, I would care for about one second, right?”— Sam

Suggested tweets

Sam Lessin: "I use a lot of Claude... But if I woke up tomorrow and Claude was gone, I would care for about one second." That's the baseline problem in the core business model.

Tweet this →

Dave Morin on the forward-deployed trap: the CFOs said "you built me this software to spend your tokens, but your tokens are really expensive."

Tweet this →

Open models vs closed models: Brit points to Vercel charts, Dave says the pie is growing too, Sam says the core premium-token business is flatlining.

Tweet this →

All topics · Full episode #169 + transcript · Subscribe on YouTube · Spotify