🟢 Welcome to the throttle economy!

Meta just decided to rate-limit smart glasses features that run entirely on device. This might be the early signal of how the throttle economy could soon reshape all business models embedding AI as a core tech layer. Let's walk through the new arithmetic that makes the decision look strange, and what could be the next throttled businesses.

Token prices as a stock-market fiction

A lean, functional CRM runs somewhere between 20,000 and 40,000 lines of code, and a native app store application of comparable ambition sits at around 25,000 lines. Now, if you ask Claude Sonnet 4.6 (Anthropic's mid-tier coding model) to write these 25,000 lines, you're in for about ten tokens of output once imports and error handling are included. So 25,000 lines land near 250,000 tokens, which at fifteen euros per million output tokens is under four dollars.

A fully functional and personalized light CRM at this price for an SME (or anyone else, really) is quite the bargain. Well, you will have to forfeit much of the software security and will probably see all your data used by Anthropic, one way or another, but hey... fifteen euros, right?

This low rate of code development is pretty much what justifies the sky-high valuations of these companies. It can't be moved (much) or challenged by reality.

But there's a catch.