Looking at popular LLM agent frameworks, a consistent pattern emerges: most systems are directed graphs with shared state, conditional routing, and feedback loops.
Adding LLM-driven decision-making and cycles turns a traditional workflow into what we now call an “agent.” The distinction feels more like a spectrum than a clear boundary.
Looking at popular LLM agent frameworks, a consistent pattern emerges: most systems are directed graphs with shared state, conditional routing, and feedback loops.
Adding LLM-driven decision-making and cycles turns a traditional workflow into what we now call an “agent.” The distinction feels more like a spectrum than a clear boundary.
Curious how others define the line in practice.