Summary & Key Terms - Agent Building Blocks

What We Learned

Build the model-tool-observation loop with validated arguments and correctly linked results.
Design tools for the model: clear names and descriptions, strict schemas, informative errors, idempotent writes.
Separate working context, durable task state and long-term memory.
Use planning and reflection where they improve task outcomes, and keep durable plans in the harness.

Term	Short meaning
Tool / function calling	A model requests a structured operation; application code executes it.
Tool-call ID	Connects a tool result to the model request that produced it.
Strict schema	Constrained decoding that guarantees tool arguments match the JSON Schema.
Tool search	Loading tool definitions on demand instead of sending every tool on every call.
Working memory / task state	Current context / persisted progress needed to continue a task.
Episodic / semantic / procedural memory	Past experiences / stored facts / reusable procedures.
Memory poisoning	Untrusted content written to memory that steers later runs.
ReAct	Reasoning and Acting: alternates reasoning, actions and observations.
Plan-and-execute	Creates a plan, executes steps and revises from feedback.
Reflection	The agent critiques its own output; helps mainly when it gets new information such as test results.