More or Less · Topics · Episode #174

Overrated, Underrated: Tune Out the AI Trade, Bet on the Application Layer

Clip from the full recording, 56:44–60:03 · download mp4 · watch at this point on YouTube

Parker says to tune out day-to-day AI trade noise, calls the application layer tremendously undervalued, and argues AI apps should earn great gross margins on token spend; Sam asks how hard it is to switch models.

Jess asks Parker for his overrated and underrated list. He says to tune out the day-to-day of the AI trade: the direction and long-term importance are clear, demand will keep growing, and there will be air pockets, but people who read stock horse-trading and conclude AI isn’t real are wrong. What’s underrated is the application layer, which he admits he has a book on.

He describes a Twitter fight over gross margins: you want to be the applied AI company with really great margins on token spend, not to prove something to investors but to yourself. If Anthropic tokens aren’t worth four times as much in your system as inside Claude, what are you doing?

Jess asks if Rippling is running open-source Chinese models. Parker says no, almost exclusively OpenAI frontier models, though they experiment with open-source. Sam asks how many lines of code it would take to flip to Anthropic or DeepSeek, and Parker says models behave differently, so you must retune evals, guidance, skills and harness, creating switching barriers.

Key points

  • Parker: tune out the day-to-day AI trade; the direction and long-term demand are clear, though there will be air pockets.
  • Parker: the application layer is tremendously undervalued.
  • Parker: you want to be the applied AI company with great gross margins on token spend, to prove it to yourself.
  • Parker: if Anthropic tokens aren't worth four times as much in your system as inside Claude, 'what are you doing?'
  • Parker: Rippling AI uses frontier models almost exclusively from OpenAI and experiments with open-source.
  • Sam: how many lines of code to flip from OpenAI to Anthropic or DeepSeek?
  • Parker: switching isn't hard but behavior differs, so evals, guidance, skills and harness need retuning.

Where they landed

ParkerIgnore the market noise; value is in applications with real margins over token cost.
SamSkeptical of switching barriers between models: how many lines of code?
JessicaAsks about overrated and underrated, then whether Rippling uses Chinese open-source models.

Quotes

“it's so clear what the direction is on this stuff and how important it's gonna be long term.”— Parker
“the application layer is like tremendously undervalued.”— Parker
“you wanna be the applied AI company with really great gross margins on token spend.”— Parker
“if anthropic tokens are not worth, you know, four X in your system than what they're worth inside of Claude, like what are you doing?”— Parker
“how many lines of code would you need to change to flip it to anthropic or to flip it to deep seek? Like two?”— Sam

Suggested tweets

Parker Conrad's underrated pick: "the application layer is like tremendously undervalued." Overrated: the day-to-day of the AI trade.

Tweet this →

Parker Conrad: if Anthropic tokens aren't worth four X in your system than inside Claude, "like what are you doing?"

Tweet this →

Sam Lessin to Parker: how many lines of code would you need to change to flip Rippling AI to Anthropic, or to DeepSeek? "Like two?"

Tweet this →

All topics · Full episode #174 + transcript · Subscribe on YouTube · Spotify