Appendix - Agent Foundations

What We Learned

• An agent chooses its next action; a workflow follows developer-defined control flow.
• Build the model-tool-observation loop with validated arguments and correctly linked results.
• Separate working context, durable task state and long-term memory.
• Use planning and human review where they improve task outcomes.
• Bound steps, time and cost; evaluate outcomes, trajectories and repeated-run reliability.

Key Acronyms, Concepts & Jargon

Term	Short meaning
Agent / workflow	Model-directed action selection / predefined execution path.
Tool / function calling	A model requests a structured operation; application code executes it.
Observation / trajectory	Result returned from an action / sequence of steps in a run.
Tool-call ID	Connects a tool result to the model request that produced it.
Working memory / task state	Current context / persisted progress needed to continue a task.
Episodic / semantic / procedural memory	Past experiences / stored facts / reusable procedures.
ReAct	Reasoning and Acting: alternates reasoning, actions and observations.
Plan-and-execute	Creates a plan, executes steps and revises from feedback.
HITL	Human In The Loop: a person reviews, corrects or approves selected steps.
Guardrail / stop condition	Checks around behavior / rule that terminates execution.
pass@k / pass^k	At least one success in k attempts / success across all k repeated trials.
Outcome / trajectory evaluation	Checks final task success / checks the actions used to reach it.

Back to section overview