In summary, even without directly banning local AI by law, you can raise the operational costs so high that it effectively functions like a ban. It's interestin

In summary, even without directly banning local AI by law, you can raise the operational costs so high that it effectively functions like a ban. It's interesting that they use Amodei as an example. And the proposed solution, making the environment surrounding the agent itself a defensive measure, is interesting. But even if state is lock-in, once a workflow is built around a specific model, whether it will work the same way with a different model is a completely different issue. That's the core problem. For example, with RAG and similar systems, companies are increasingly using AI agents internally. But building a RAG system inevitably depends on the specific model being used, and I'm not sure if that's really solvable.