Author here. TL;DR: LLMs made language a UI, but “real work” hits two limits: context rot + accountability. I propose replacing “agent” with 4 modes (Producer/Retriever/Operator/Judge) and using 5 questions before shipping outputs.
Would love critique on:
What’s the weakest part of the argument?
What real-world counterexample breaks this framework?
Author here. TL;DR: LLMs made language a UI, but “real work” hits two limits: context rot + accountability. I propose replacing “agent” with 4 modes (Producer/Retriever/Operator/Judge) and using 5 questions before shipping outputs. Would love critique on:
What’s the weakest part of the argument? What real-world counterexample breaks this framework?