Agentic AI Notebook
0What is an AI Agent?1Why LLMs Need Agents2Anatomy of an Agent3Agent Lifecycle4Core Concepts5Agent Capabilities6Types of Agents7Agent Architectures8Agent Terminology9Current Agent Landscape10Planning11Reflection12Multi Tool13Self Correction14Build First AI Agent
Agent Foundations
Phase 4Module 9 of 15

Agent Terminology

Precise terms prevent miscommunication in teams, docs, and incident reports.

Like learning repo, PR, and CI before contributing to a codebase.

Visual Workflows

Start here — scroll inside each diagram frame to explore, then use + / − to zoom up to 200% if needed.

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

Agent Loop Vocabulary

100%
Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

Terms you will see in traces logs and incident reports.

Key Takeaways

  • 1.Agent: LLM plus tools in autonomous loop.
  • 2.Trajectory: full thought-action-observation sequence.
  • 3.Episode: one task from start to finish.
  • 4.Guardrails: safety filters on input and output.

Real Example

Scenario

Incident P0: Episode `ticket-8821` had trajectory [search×3 failed → cache fallback → answer]. Termination reason was `max_steps`, not `done`.

What you would do

Episode = one support ticket end-to-end. Trajectory = ordered action/observation log. Three failed search actions mean the agent never grounded on live data. Termination `max_steps` = budget exhausted without success. Add a golden eval for triple-search failure and document the cache fallback policy.

Practice Task

Write a fake 5-line trajectory for a refund lookup. Label: episode ID, each action, each observation, and the termination reason.

Code Walkthrough

Highlighted lines show where Agent Terminology happens in the code.

Agent Terminology
1from dataclasses import dataclass, field  # import dependencies2
3@dataclass4class TrajectoryStep:  # define a data structure or component5    action: str6    observation: str7
8@dataclass9class Episode:  # define a data structure or component10    episode_id: str11    steps: list[TrajectoryStep] = field(default_factory=list)12    termination: str = "running"13
14ep = Episode(episode_id="ticket-8821")15ep.steps.append(TrajectoryStep("web_search", "0 results"))  # key line for Agent Terminology16ep.steps.append(TrajectoryStep("web_search", "0 results"))17ep.termination = "max_steps"18print(ep.episode_id, len(ep.steps), ep.termination)  # show output for debugging

Cheat Sheet

Quick recap

quick ref
  • •Trajectory = full debug log
  • •Episode = one complete task
  • •Observation = tool result
  • •Grounding = cite sources
  • •Eval = regression test suite

Common Mistakes

  • ✕Skipping evaluation for Agent Terminology before production
  • ✕No logging or tracing around agent terminology steps
  • ✕Ignoring cost and latency implications