saved
The Big Discount on Intelligence
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
Arunabh Dastidar argues that AI subscriptions sell intelligence far below API prices: moving his five-engineer team from Max and Pro plans to the API raised costs about thirty times. He reads the subsidy as payment in decision traces captured by lab harnesses, and urges founders to take the discount while owning their knowledge and loops.
ideas
- The subsidy is huge. Dastidar estimates a maxed-out $200 plan costs $4,000 to $6,000 at API prices, and cites outside estimates of deeply negative margins on heavy subscribers.
- Subscribers are unpaid trainers. Labs pay RL-environment vendors billions for signal, while active subscribers supply thousands of corrected examples a month for a flat fee.
- The best models stay inside. Restricted systems like OpenAI's math model and Claude Mythos Preview suggest labs keep their strongest intelligence as a competitive edge.
- Doing the work beats selling seats. Anthropic's reported $30 trillion market claim points to AI-native competitors in finance, retail, and trading built on unreleased models.
- Own your decisions. Harnesses see how work is done even when enterprise terms exclude training, so decision logs, workflows, and verified signal should stay under your control.
quotes
“After this, the bill was approximately $1000 per day for the same work. So in one fell swoop, the cost went up thirty times.”
“Once the AI exists, why sell it at a hundredfold discount?”
“Selling seats to customers will not take a company to $30 trillion. Doing the work might.”
“The record of how your people make decisions is the asset.”