“If the energy it requires to send somebody to RevenueCat or Expo or whatever is always less than the energy it would take to generate the tokens, like the model should just optimize towards that outcome.”

The energy-gradient: AI keeps recommending the same tools because it's cheaper than generating from scratch

Jacob's gradient-descent framing: models will route users to existing solutions whenever that's cheaper than generating equivalent code. Once you're the established 'relative minimum' for a problem (auth, payments, mobile delivery), the model keeps reinforcing the referral. New entrants only get a look where there's genuinely unexplored space.