When you think so hard for how to build agents, most likely agents don't have a solid usecase beyond being a glorified chat bot plus the traditional automation workflows.

Let's be honest. What you call as an agent is the glue between the human user, tools and LLMs. This glue exists solely to allow the humans use their natural language. If you remove the need for NLP, there is no agent.

You might say agents take actions, define their workflows on the fly, does activities autonomously etc. Again, just fluff. Nobody would ever let a bot define workflow as workflows are tied to business rules and processes. These rules and processes are not to be "defined" on the fly. Autonomous activities? Automations are nothing new.

The only new thing in the town is ability to convert the human-like like language into a structured information. Once this is done, everything continues on traditional rails. These AI layer is just skin deep, giving humans the comfort of using their natural language to interact with computers. If anyone says anything else, they are priming the pumps.