01
Agent Trajectory
Definition
The full path of an agent run, including plans, reasoning summaries, tool calls, observations, retries, and final output.
Back to Agentic AI Glossary
Agentic AI Glossary
Agentic AI Foundations & Architectures
Agentic AI Foundations & Architectures terms and explanations from the Agentic AI Glossary.
131 terms in this chapter
02
Critique Loop
Definition
Critique Loop is a repeated control cycle for critique loop. It lets an agent adjust its next step after reasoning, taking action, observing results, or receiving critique.
03
Error Recovery
Definition
The agent's ability to detect a failure, choose a safe fallback, retry carefully, or ask for help without losing the task.
04
Final Response
Definition
The user-facing answer or action summary produced after planning, retrieval, tool calls, and validation are complete.
05
Intermediate Step
Definition
A single internal action in an agent run, such as planning, retrieving, calling a tool, checking output, or revising state.
06
Iteration Limit
Definition
A hard cap on the number of reasoning or action cycles allowed, preventing infinite loops and runaway cost.
07
Loop Termination
Definition
Loop Termination is a repeated control cycle for loop termination. It lets an agent adjust its next step after reasoning, taking action, observing results, or receiving critique.
08
Max Turns
Definition
The maximum number of back-and-forth agent steps or conversation turns allowed before stopping, escalating, or returning a result.
09
Plan-Act-Observe
Definition
An agent loop where the system creates a plan, executes an action, observes the result, and decides the next step.
10
ReAct Loop
Definition
ReAct Loop is a repeated control cycle for react loop. It lets an agent adjust its next step after reasoning, taking action, observing results, or receiving critique.
11
Reason-Act-Observe
Definition
A loop pattern where the model reasons about the next move, performs an action, then uses the observation to continue.
12
Reflection Loop
Definition
Reflection Loop is a repeated control cycle for reflection loop. It lets an agent adjust its next step after reasoning, taking action, observing results, or receiving critique.
13
Replanning
Definition
Updating the original plan when tools fail, new information appears, constraints change, or the current path is no longer valid.
14
Retry Loop
Definition
Retry Loop is a repeated control cycle for retry loop. It lets an agent adjust its next step after reasoning, taking action, observing results, or receiving critique.
15
Scratchpad
Definition
A private working area used to store intermediate notes, state, tool results, or structured reasoning summaries during a run.
16
Self-Correction
Definition
The agent's process of identifying mistakes in its own output or actions and revising them before final delivery.
17
Sense-Think-Act
Definition
A classic agent loop: perceive the environment, decide what the information means, then take an appropriate action.
18
State Update
Definition
A change to stored task status, memory, variables, or workflow progress after an observation or action.
19
Thought
Definition
An internal reasoning unit or planning note used by some agent frameworks to organize decisions before action.
20
Verifier Loop
Definition
Verifier Loop is a repeated control cycle for verifier loop. It lets an agent adjust its next step after reasoning, taking action, observing results, or receiving critique.
21
Working Memory
Definition
Working Memory is the active information an agent uses while solving the current task, such as the plan, recent observations, and pending steps.
22
Acting
Definition
The execution part of an agent loop where the agent does something, such as calling a tool or sending a response.
23
Action
Definition
The execution part of an agent loop where the agent does something, such as calling a tool or sending a response In this book, it is defined specifically for Action within Agentic AI Foundations.
24
Action Space
Definition
The set of actions, tools, APIs, or workflows an agent is allowed to execute.
25
Agency
Definition
Agency is a foundation concept in agentic AI. It explains how agents handle agency while pursuing goals, making decisions, acting, or staying controlled.
26
Agent Loop
Definition
Agent Loop is a foundation concept in agentic AI. It explains how agents handle agent loop while pursuing goals, making decisions, acting, or staying controlled.
27
Agentic AI
Definition
AI systems that pursue goals through planning, tool use, memory, observation, and iterative action rather than only generating one response.
28
AI Agent
Definition
A software component powered by an AI model that can interpret context, choose actions, call tools, and work toward a defined objective.
29
Autonomous Agent
Definition
An Autonomous Agent can decide and act across multiple steps with limited human intervention while staying inside defined boundaries.
30
Autonomy
Definition
The degree of independence an agent has when deciding, acting, and recovering without direct human intervention.
31
Completion Criteria
Definition
Completion Criteria is a foundation concept in agentic AI. It explains how agents handle completion criteria while pursuing goals, making decisions, acting, or staying controlled.
32
Environment
Definition
Environment is a foundation concept in agentic AI. It explains how agents handle environment while pursuing goals, making decisions, acting, or staying controlled.
33
Executor
Definition
The component or agent role that performs planned steps by calling tools, APIs, or internal workflows.
34
Feedback
Definition
Feedback is a foundation concept in agentic AI. It explains how agents handle feedback while pursuing goals, making decisions, acting, or staying controlled.
35
Goal
Definition
The target outcome an agent is trying to reach through planning and action.
36
Goal-Oriented Agent
Definition
A Goal-Oriented Agent works toward a defined outcome instead of only answering one isolated prompt.
37
Goal-Oriented Behavior
Definition
The ability to optimize actions around an outcome instead of treating every request as an isolated message.
38
Human-in-the-Loop
Definition
A pattern where humans review, approve, correct, or complete parts of an agent workflow.
39
Human-on-the-Loop
Definition
A supervision pattern where humans monitor agent behavior and intervene when thresholds or risks appear.
40
Human-out-of-the-Loop
Definition
Human-out-of-the-Loop is a foundation concept in agentic AI. It explains how agents handle human-out-of-the-loop while pursuing goals, making decisions, acting, or staying controlled.
41
Memory
Definition
The storage and retrieval layer that lets an agent preserve useful information across turns, sessions, or tasks.
42
Multi-Agent System
Definition
A design where multiple specialized agents coordinate, delegate, review, or debate to complete a complex task.
43
Observation
Definition
Information returned after an agent acts, such as a tool result, error message, retrieved document, or user response.
44
Perception
Definition
The process of receiving and interpreting input from users, tools, documents, events, or the environment.
45
Planner
Definition
The agent component that decomposes a goal into steps, dependencies, and execution order.
46
Planning
Definition
Planning is a foundation concept in agentic AI. It explains how agents handle planning while pursuing goals, making decisions, acting, or staying controlled.
47
Policy
Definition
Rules and constraints that guide how an agent should decide, respond, escalate, or use tools.
48
Reasoning
Definition
Reasoning is a foundation concept in agentic AI. It explains how agents handle reasoning while pursuing goals, making decisions, acting, or staying controlled.
49
Reasoning Engine
Definition
The decision layer that evaluates context, constraints, options, and evidence before choosing the next step.
50
Reflection
Definition
An agent pattern where the system reviews its own plan, answer, or tool trajectory and attempts improvement.
51
Semi-Autonomous Agent
Definition
A Semi-Autonomous Agent handles routine steps by itself but asks for human approval when risk, uncertainty, or policy limits appear.
52
Single-Agent System
Definition
Single-Agent System is a foundation concept in agentic AI. It explains how agents handle single-agent system while pursuing goals, making decisions, acting, or staying controlled.
53
State
Definition
The current snapshot of an agent workflow, including context, progress, memory, and pending decisions.
54
Step
Definition
One unit of work inside a larger agent plan or execution sequence.
55
Task
Definition
A specific piece of work the agent is asked to complete or contribute to.
56
Task-Oriented Agent
Definition
A Task-Oriented Agent is specialized for a specific type of work, such as scheduling, support, testing, coding, or data analysis.
57
Tool Use
Definition
Tool Use is a foundation concept in agentic AI. It explains how agents handle tool use while pursuing goals, making decisions, acting, or staying controlled.
58
Activations
Definition
Intermediate values produced inside a neural network after applying transformations to inputs.
59
Artificial Intelligence
Definition
The field of building systems that can perceive, reason, learn, generate, or act in ways associated with human intelligence.
60
Attention
Definition
A mechanism that lets a model focus on the most relevant parts of input context.
61
Backpropagation
Definition
The algorithm used to compute gradients through a neural network during training.
62
Bias
Definition
Systematic error, preference, or imbalance in model behavior, data, or evaluation.
63
Decoder
Definition
A model component that generates output tokens from internal representations and prior tokens.
64
Deep Learning
Definition
Machine learning based on multi-layer neural networks that learn hierarchical representations from data.
65
Encoder
Definition
A model component that converts input into internal representations.
66
Evaluation
Definition
Measuring model or system quality against desired criteria such as correctness, safety, and business value.
67
Fine-Tuning
Definition
Additional training that adapts a model to a task, domain, style, or preference set.
68
Foundation Model
Definition
A general-purpose model trained at scale and adapted to tasks through prompting, retrieval, fine-tuning, or tool use.
69
Generalization
Definition
A model's ability to perform well on new examples beyond its training data.
70
Generative AI
Definition
AI that creates new content such as text, code, images, audio, plans, or structured data.
71
Gradient Descent
Definition
An optimization method that updates parameters in the direction that reduces loss.
72
Inference
Definition
Running a trained model to produce predictions, answers, actions, or generated content.
73
Large Language Model
Definition
A neural language model trained on large text datasets and used to understand, generate, summarize, reason, and follow instructions.
74
Loss Function
Definition
A mathematical objective that measures how wrong a model is during training.
75
Machine Learning
Definition
A branch of AI where models learn patterns from data instead of being explicitly programmed for every rule.
76
Multimodal Model
Definition
Multimodal Model is a model type or component focused on multimodal. It helps produce, transform, rank, interpret, or evaluate AI outputs.
77
Neural Network
Definition
A model composed of connected layers that transform input data into predictions or generated outputs.
78
Optimization
Definition
Improving a model, prompt, workflow, or system to meet accuracy, cost, speed, or reliability goals.
79
Overfitting
Definition
When a model memorizes training data patterns and performs poorly on new data.
80
Parameters
Definition
Learned numerical values inside a model that determine how it transforms inputs into outputs.
81
Pretraining
Definition
Large-scale initial model training on broad data before task-specific adaptation.
82
Self-Attention
Definition
Attention applied within the same sequence so each token can relate to other tokens.
83
Sequence-to-Sequence Model
Definition
Sequence-to-Sequence Model is a model type or component focused on sequence-to-sequence. It helps produce, transform, rank, interpret, or evaluate AI outputs.
84
Small Language Model
Definition
Small Language Model is a model type or component focused on small language. It helps produce, transform, rank, interpret, or evaluate AI outputs.
85
Training
Definition
The process of updating model parameters using data and an optimization objective.
86
Transformer
Definition
A neural architecture based on attention mechanisms that powers most modern LLMs.
87
Underfitting
Definition
When a model is too simple or poorly trained to capture useful patterns.
88
Variance
Definition
Sensitivity of a model to changes in training data or input conditions.
89
Weights
Definition
Model parameters that control the strength of connections between neural network components.
90
Accessible Environment
Definition
An environment where the agent's sensors can access the full state needed to choose actions.
91
Actuator
Definition
A physical or software mechanism that turns an agent's decision into an action in the environment.
92
Agent
Definition
A system or software program that perceives an environment and acts toward one or more goals.
93
Agent Function
Definition
The mapping from an agent's percept history to the action it should take next.
94
Agent Program
Definition
The implementation that runs on an architecture and produces actions from percepts or state.
95
Behavior of Agent
Definition
The actions an agent performs after receiving a particular sequence of percepts.
96
Condition-Action Rule
Definition
A rule that maps a detected condition directly to an action, often used in reflex agents.
97
Continuous Environment
Definition
An environment whose states or actions can vary across a continuous range rather than fixed options.
98
Deterministic Environment
Definition
An environment where the next state is fully determined by the current state and the agent's action.
99
Discrete Environment
Definition
An environment with a limited set of clearly separated states, actions, or time steps.
100
Dynamic Environment
Definition
An environment that can change while an agent is deciding or acting.
101
Effector
Definition
A body part, motor, actuator, or software output channel that carries out an agent's action.
102
Episodic Environment
Definition
An environment where each decision episode is independent of previous episodes.
103
Goal-Based Agent
Definition
An agent that chooses actions by considering which steps will move it closer to a desired goal.
104
Human Agent
Definition
A human viewed as an agent with sensory organs for perception and body parts for action.
105
Ideal Rational Agent
Definition
An agent that selects the action expected to maximize its performance measure using percepts and knowledge.
106
Inaccessible Environment
Definition
An environment where the agent cannot directly sense all information needed for a complete state description.
107
Internal State
Definition
The agent's stored representation of aspects of the world that are not currently observable.
108
Model-Based Reflex Agent
Definition
A reflex agent that maintains an internal model of how the world changes and how actions affect it.
109
Multi-Agent Environment
Definition
An environment containing more than one agent, where agents may cooperate, compete, or affect each other.
110
Non-Deterministic Environment
Definition
An environment where an action may lead to different possible next states even from the same current state.
111
Non-Episodic Environment
Definition
An environment where current actions can affect later situations and future rewards.
112
Observable Environment
Definition
An environment where percepts reveal the full current state relevant to the agent's decision.
113
Partially Observable Environment
Definition
An environment where the agent receives incomplete information and must reason with uncertainty.
114
PEAS
Definition
A task specification using Performance measure, Environment, Actuators, and Sensors to describe an agent problem.
115
Percept
Definition
The information an agent receives from its sensors at a particular moment.
116
Percept Sequence
Definition
The complete history of percepts an agent has received so far.
117
Percepts
Definition
The sensed inputs an agent receives from its environment.
118
Performance Measure of Agent
Definition
The success criteria used to judge how well an agent is performing its task.
119
Rational Agent
Definition
An agent that chooses actions expected to achieve the best outcome based on its knowledge and percepts.
120
Rationality
Definition
The quality of choosing sensible actions that are expected to produce useful results.
121
Robotic Agent
Definition
A physical agent that uses sensors such as cameras and effectors such as motors to act in the real world.
122
Sensor
Definition
A device or software input channel that lets an agent detect information about its environment.
123
Simple Reflex Agent
Definition
An agent that chooses actions using only the current percept and condition-action rules.
124
Single-Agent Environment
Definition
An environment in which one agent acts without strategic interaction with other agents.
125
Softbot
Definition
A software robot that acts inside digital environments such as websites, file systems, or databases.
126
Software Agent
Definition
An agent implemented as software that acts through encoded instructions, APIs, tools, or digital actions.
127
Static Environment
Definition
An environment that does not change while an agent is deciding what to do.
128
Turing Test Environment
Definition
A test setting where a human judge compares typed responses from a human and a machine.
129
Utility
Definition
A numeric preference score that represents how desirable a state or outcome is for an agent.
130
Utility-Based Agent
Definition
An agent that chooses actions by comparing the expected utility of possible outcomes.
131
World Model
Definition
An agent's representation of how the environment behaves and how actions change it.
Explore more chapters or test your knowledge with quizzes.