In this workshop we're going to adjust the amount of power we give to the LLM to choose our control flow.

Doing this bumps up against a hot topic in AI engineering circles - agents vs workflows.

The opposite of an 'agent' is a 'workflow'.

The less 'agentic' your application is, the more predictable it becomes. Entirely deterministic applications do the same thing every time - even if they contain some probabilistic components.

Agentic systems aren't all bad - letting the LLM handle the control flow means your application becomes very flexible.

But in a system like DeepSearch, we may want to experiment with adding more determinism to our system.

In our next few exercises, we're going to experiment with turning the dial back towards determinism.