Part II — Agents

Distinguishes an LLM, a chatbot, a tool-using LLM, and an agent, and introduces the Reason → Act → Observe → Repeat loop with non-financial examples.

Short-term (conversation) memory versus long-term (user) memory versus state management, their practical limits, and a demonstration in Agno.

The tool contract -- Input / Processing / Output / Validation -- that bridges a Python function to something an agent can call, including tool selection and validation via RetryAgentRun.

Model selection, temperature and sampling and their effect on tool-selection consistency, system instructions, and structured-output configuration -- closing on why consistency is not the same as correctness.

The M×N integration problem, MCP's host / client / server architecture and its tools / resources / prompts primitives, and connecting an Agno agent to a real MCP server.

Why and how to split a broad tool set across specialist agents, Agno's Team (route / coordinate / broadcast modes) and Workflow, and the trade-offs against a single agent.