X @Balaji

I understand these points but disagree because:(1) Many/most of these agents are using the same upstream model (namely Claude) so it’s just Claude talking to itself.(2) The longer these agents talk to themselves past a prompt, the more they get off track and lose coherence. They are built for the leash, and need constant direction and guidance.(3) Unpredictability of an AI agent acting on your behalf is a bug, not a feature. There are many ways for things to go unpredictably wrong and very few for them to g ...