this feels anti-trust-y to me.

when i signed up for a subscription it was with the understanding that id be able to use those tokens on which ever agent i wanted to play with, and that as i got to something i want to have persistently running, id switch that to be an api client. i quickly figured out that claude code was the current best coding agent for the model, but seeing other folks calling opus now im not actually sure thats true, in which case that subsidized token might be more expensive to both me and anthropic, because its not the most token efficient route over their model.

i dislike that now i wont be able to feed them training data using many different starting points and paths, which i think over time will have a bad impact on their models making them worse over time