The Dispatch Problem
A customer support system at a fintech startup gets this message: “My transfer failed yesterday, can you check the status and also tell me if my tier qualifies for waived fees?”
A dumb LLM call answers one question and halts. A well-built agent understands this requires two tool calls (transaction lookup, account tier check), sequencing them based on partial results, remembering context from earlier in the conversation, and producing a single coherent reply.
The difference between these two outcomes is not the model — it’s the agent framework wrapped around it. Understanding how memory, tools, and planning modules fit together is the foundational skill for anyone building production AI systems in 2025 and beyond.
Core Concept: The Three-Module Architecture
An agent framework is a runtime loop that connects an LLM to the outside world through three distinct subsystems.
