AI Agents, Hype or Snakeoil?

Trying to understand the hype around AI agents. It's clear the AI is not intelligent enough to autonomously complete tasks, thus it needs to be encoded with some kind of decision tree. This seems to defeat the purpose of leveraging AI's intelligence. People also seem to be excited about LLMs automation of using websites, but these application would likely run into the same things that current automation runs into(rate limits, bot detection etc).