Adversarial Machine Learning
Adversarial Machine Learning is the field that studies how Machine Learning systems can be deceived or attacked with deliberately crafted data, while also developing measures to detect and defend against such attacks.
AI Glossary
AI Glossary
100 AI and Agentic AI terms — explained in plain language
⌘K
Popular:
100 terms · Updated Sep 2026
A
33 terms
Agent Architecture
Agent Architecture is the way the components inside an AI Agent are organized, including the AI model, memory, data, tools, and rules, so that the Agent can coordinate these components when performing tasks.
Agent Autonomy
Agent Autonomy is the degree to which an AI Agent can plan, make decisions, and take actions on its own without needing human guidance or approval at every step.
Agent Evaluation
Agent Evaluation is the process of assessing how well an AI Agent achieves its goals, uses the right tools, follows rules, and produces accurate, consistent results in real-world situations.
Agent Handoff
Agent Handoff is the mechanism for transferring a task and all related information from one AI Agent to another Agent or to a human, so the work can be continued by the party best able to handle it.
Agent Long-Term Memory
Agent Long-Term Memory is the memory that lets an AI Agent retain knowledge, interaction history, or preferences across multiple sessions so it can reuse them in future tasks.
Agent Loop
Agent Loop is the cycle in which an AI Agent continuously takes in information, chooses an action, checks the result, and adjusts its next step until the task is completed or must be stopped.
Agent Memory
Agent Memory is the mechanism that lets an AI Agent store and reuse information from previous conversations, data, or actions to maintain context and support decision-making.
Agent Observability
Agent Observability is the ability to monitor how an AI Agent operates, including the steps it has taken, the tools it has used, the errors that occurred, processing time, cost, and the final result.
Agent Planning
Agent Planning is the ability of an AI Agent to identify intermediate goals, sequence its actions, and update its plan when new information or results emerge.
Agent Reflection
Agent Reflection is the mechanism that lets an AI Agent review its own results and process, recognize shortcomings, and then adjust its plan or approach on the next attempt.
Agent Short-Term Memory
Agent Short-Term Memory is temporary memory that helps an AI Agent keep track of the request, data, and steps already taken in the current task. This information is usually not retained long-term after the session ends.
Agent Task Decomposition
Agent Task Decomposition is the process by which an AI Agent breaks a large, complex task into smaller tasks and then determines the order of, or relationships between, the steps so they can be carried out more effectively.
Agent2Agent Protocol (A2A)
Agent2Agent Protocol (A2A) is an open standard that enables AI Agents built on different platforms to discover each other's capabilities, exchange information, delegate tasks, and coordinate with one another.
Agentic AI
Agentic AI is a form of AI that can plan, use tools, and adjust its approach based on the results it receives in order to achieve a goal with less human guidance.
Agentic RAG
Agentic RAG is an implementation of RAG in which an AI Agent actively decides what information to look for, selects appropriate sources, performs multiple rounds of searching, and checks the evidence before answering.
Agentic Workflow
Agentic Workflow is a way of organizing work in which an AI Agent can determine the order of execution, use tools, and adjust its actions based on actual results instead of simply following a fixed script.
AgentOps
AgentOps is the set of methods and tools used to deploy, monitor, evaluate, manage, and improve AI Agents in real-world production environments.
AI Agent
AI Agent is an AI system that is given a goal and can determine the steps needed, use tools, and adjust its actions to complete the task within the scope of the permissions it has been granted.
AI Alignment
AI Alignment is the process of adjusting the goals and behavior of AI so that the system acts in line with human intentions, values, safety principles, and interests.
AI Bias
AI Bias is the phenomenon in which AI produces skewed or unfair results due to problems in the data, the model's design, or the way the system is used in practice.
AI Ethics
AI Ethics is the field of principles and standards that guide the development and use of AI in a way that is transparent, fair, safe, responsible, and respectful of privacy.
AI Governance
AI Governance is the system of policies, roles, and processes that helps an organization control how AI is developed and used, ensuring alignment with its goals, the law, safety requirements, and accountability.
AI Grounding
AI Grounding is the process of linking an AI's answers or actions to reliable data, documents, or sources of evidence to improve accuracy and verifiability.
AI Guardrails
AI Guardrails are a set of rules and control layers that limit how AI responds or acts, in order to prevent inappropriate, inaccurate, or risky content.
AI Inference
AI Inference is the process of using a trained model to generate answers, predictions, or decisions from new data without retraining the model.
AI Orchestration
AI Orchestration is the process of coordinating multiple AI models, AI Agents, tools, data, and rules within a unified workflow so that the components work together in the right order and achieve a shared goal.
AI Safety
AI Safety is the field focused on identifying, assessing, and mitigating the risks that AI may pose to people, organizations, or systems during its development and use.
AI-Driven Personalization
AI-Driven Personalization is the approach of using AI to analyze each individual's behavior, preferences, and interaction history, and then tailor content, products, offers, or experiences to better suit them.
API (Application Programming Interface)
API (Application Programming Interface) is an interface that allows software programs to exchange data and requests with each other. Through APIs, AI systems can connect to CRMs, data warehouses, e-commerce marketplaces, or external services.
Artificial General Intelligence (AGI)
Artificial General Intelligence (AGI) is the concept of an AI system that can learn, reason, and adapt across many kinds of intellectual tasks as humans do, rather than excelling only in one specific domain. AGI is not currently recognized as having been achieved.
Attention Mechanism
Attention Mechanism is a mechanism that helps an AI model identify the most important parts of the information in the data, so it can focus on processing the relevant relationships when generating results.
Autonomous Agent
Autonomous Agent is an AI Agent that can proactively pursue goals, break down tasks, choose tools, and adjust its actions without detailed human guidance at every step.
B
1 terms
Big Data
Big Data is a concept referring to datasets whose size, speed of generation, or variety is so large that specialized technologies are needed to store, process, and analyze them.
C
8 terms
Chain-of-Thought Prompting
Chain-of-Thought Prompting is a technique that guides an AI model to solve a problem through intermediate reasoning steps, and is often used for tasks that require multi-step analysis, calculation, or planning.
Cloud AI
Cloud AI is AI services and infrastructure delivered through cloud computing platforms, allowing businesses to train, deploy, or use models without investing in an entire hardware system themselves.
Computer Vision
Computer Vision is the field of AI that enables computers to recognize and analyze content in images or videos, such as classifying images, detecting objects, reading text, or tracking motion.
Constitutional AI
Constitutional AI is a method of aligning AI based on a predefined set of principles, enabling the model to evaluate and revise its own responses to better comply with those principles.
Context Engineering
Context Engineering is the process of selecting and organizing all the information a model or AI Agent needs to perform a task, including instructions, data, memory, tools, and the current state.
Context Window
Context Window is the maximum amount of information an AI model can consider at one time, including instructions, conversation history, and provided documents. It is not the AI's long-term memory.
Continual Learning
Continual Learning is a method that enables an AI model to keep learning from new data over time without losing most of the important knowledge it learned previously.
Conversational Commerce
Conversational Commerce is a form of buying, selling, and customer care conducted through conversations on messaging channels, where businesses or AI can advise customers, recommend products, and support transactions.
D
4 terms
Data Mapping
Data Mapping is the process of identifying and linking corresponding data fields across multiple sources or systems, so that information about customers, products, and orders is converted consistently.
Deep Learning
Deep Learning is a Machine Learning method that uses multi-layer neural networks to learn complex features from large amounts of data such as text, images, audio, or video.
Diffusion Model
Diffusion Model is an AI model that generates content by starting from noisy data and gradually reconstructing it into a structured result. This method is commonly used to generate images, audio, or video.
Digital Expert
Digital Expert is a term describing a specialized AI system equipped with the knowledge, skills, and tools for a business domain such as data analysis, planning, or content creation.
E
4 terms
E-commerce Automation
E-commerce Automation is the process of using AI and software to automate repetitive e-commerce tasks such as replying to comments, advising on products, updating orders, and providing customer care.
Embedding
Embedding is the way AI converts text, images, or data into sequences of numbers in order to compare meaning, measure similarity, and find content that is related.
Episodic Memory
Episodic Memory is a type of memory that lets an AI Agent record events and experiences that have occurred, so it can refer to how it handled things before when it encounters similar situations.
Explainable AI (XAI)
Explainable AI (XAI) is a set of methods that help people understand the key factors or data that influenced the predictions, recommendations, or decisions of an AI system.
F
5 terms
Federated Learning
Federated Learning is a method of training AI across multiple devices or data sources without needing to move all of the original data to a central storage system.
Few-shot Prompting
Few-shot Prompting is a technique of providing a few sample examples directly in the prompt so the AI model can recognize the desired structure, patterns, and way of answering without retraining.
Fine-tuning
Fine-tuning is the process of further training an existing AI model with specialized data, making the model better suited to a specific task, domain, communication style, or organizational need.
Foundation Model
Foundation Model is an AI model trained on large, diverse amounts of data to serve as the foundation for many different applications. The model can be adapted to serve specific tasks or domains.
Function Calling
Function Calling is a mechanism that allows an AI model to generate a structured request containing a function name and the necessary information, so the application can validate it and then execute the corresponding function.
G
1 terms
Generative AI
Generative AI is a type of AI that can create new content such as text, images, audio, video, or source code by learning patterns and relationships from the data it was trained on.
H
3 terms
Hallucination
Hallucination is the phenomenon in which AI generates information that sounds plausible but is actually inaccurate, unfounded, or nonexistent. RAG and Grounding can reduce this risk but cannot eliminate it entirely.
Human-AI Collaboration
Human-AI Collaboration is a way of working in which AI helps analyze, process, or automate tasks, while humans set the goals, oversee risks, handle exceptions, and take responsibility for the final decision.
Human-in-the-Loop (HITL)
Human-in-the-Loop (HITL) is a way of operating AI with human involvement at critical steps, such as reviewing results, handling exceptions, or approving high-risk actions.
K
1 terms
Knowledge Base
Knowledge Base is a repository of documents, data, and information organized so that people or AI systems can search, consult, and use it when answering questions.
L
3 terms
Lead Scoring
Lead Scoring is a method of evaluating and ranking prospective customers based on their information, behavior, and level of engagement, helping sales teams prioritize the opportunities most likely to convert.
LLM (Large Language Model)
LLM (Large Language Model) is an AI model trained on large amounts of text data to understand and generate language, enabling it to answer questions, write content, summarize, or assist with analysis.
LLM Token
LLM Token is the basic unit of information that a language model uses to read and generate text. A token can be a word, part of a word, a character, or a punctuation mark.
M
8 terms
Machine Learning
Machine Learning is a branch of AI that enables systems to learn patterns from data in order to make predictions or decisions without humans programming every case in detail.
Mixture of Experts (MoE)
Mixture of Experts (MoE) is a way of building an AI model with multiple specialized components. When it receives a request, the system uses only the most suitable components, improving efficiency and saving computing resources.
Model Context Protocol (MCP)
Model Context Protocol (MCP) is a standard that lets AI applications connect to external data, tools, and systems in a unified way, instead of having to build a separate integration for each source.
Model Distillation
Model Distillation is a technique for transferring part of the knowledge and capabilities of a large AI model to a smaller one, reducing cost and processing time while still retaining the necessary capabilities.
Model Drift
Model Drift is the phenomenon in which an AI model's performance declines over time as data, user behavior, or the real-world environment changes compared with when the model was trained.
Model Routing
Model Routing is the mechanism for selecting the appropriate AI model for each request based on factors such as difficulty, processing capability, speed, cost, privacy, and desired quality.
Multi-Agent System
Multi-Agent System is a system of multiple AI Agents with different roles or capabilities that exchange information, divide up work, and coordinate to solve a shared task.
Multimodal AI
Multimodal AI is a type of AI that can take in, combine, and process multiple forms of information, such as text, images, audio, or video, within the same task.
N
2 terms
Natural Language Generation (NLG)
Natural Language Generation (NLG) is a technology that enables AI to turn data or structured information into easy-to-understand text, such as product descriptions, reports, notifications, or automated answers.
Natural Language Processing (NLP)
Natural Language Processing (NLP) is the field of AI that enables computers to analyze, understand, and generate human language, providing the foundation for AI assistants, semantic search, sentiment analysis, and automated conversations.
O
1 terms
Omnichannel AI
Omnichannel AI is the use of AI to unify and handle interactions from multiple channels such as Website, Facebook, Zalo, or e-commerce marketplaces, ensuring consistent responses and maintaining context across channels.
P
5 terms
Predictive AI
Predictive AI is a type of AI that uses historical data to forecast future trends, demand, behavior, or risks, helping people make data-driven decisions.
Privacy-Preserving AI
Privacy-Preserving AI is a set of methods that allow AI to learn from or process data while limiting the risk of disclosure, re-identification, or unauthorized use of sensitive information.
Prompt Chaining
Prompt Chaining is a technique that links multiple prompts into a multi-step process, in which the output of each step is used as data or context for the next step.
Prompt Engineering
Prompt Engineering is the process of designing, testing, and refining input instructions so that an AI model correctly understands the task, uses sufficient information, and produces results that fit the goal.
Prompt Injection
Prompt Injection is a form of attack that inserts malicious instructions or misleading content into input data in order to make AI ignore its rules, disclose information, or perform unintended actions.
R
8 terms
ReAct (Reasoning and Acting)
ReAct (Reasoning and Acting) is a method that helps an AI Agent combine analysis with action. After each action, the Agent observes the result and uses the new information to decide the next step.
Reasoning Model
Reasoning Model is an AI model optimized for tasks that require multi-step analysis, weighing conditions, and checking results before giving an answer or making a decision.
Reinforcement Learning (RL)
Reinforcement Learning (RL) is a Machine Learning method in which AI learns to choose actions through the rewards or penalties it receives while interacting with an environment over time.
Reinforcement Learning from Human Feedback (RLHF)
Reinforcement Learning from Human Feedback (RLHF) is a technique for adjusting an AI model based on human ratings, rankings, or choices, making its results better aligned with user expectations and safety principles.
Responsible AI
Responsible AI is an approach to developing and using AI based on the principles of fairness, transparency, safety, privacy protection, explainability, and accountability.
Retrieval-Augmented Generation (RAG)
Retrieval-Augmented Generation (RAG) is an approach that combines an AI model with a step that retrieves information from specified documents or data sources before generating an answer, making the results better grounded and more consistent with real data.
Robotic Process Automation (RPA)
Robotic Process Automation (RPA) is a technology that uses software robots to automatically perform rule-based operations on digital systems. RPA does not necessarily use AI but can be combined with AI.
Router Agent
Router Agent is an AI Agent that specializes in analyzing requests and routing tasks to the most suitable model, tool, or Agent based on capability, cost, speed, and established rules.
S
9 terms
Scalability
Scalability is the ability of an AI system to expand to handle growing numbers of users, data, or requests while keeping performance, stability, cost, and service quality at appropriate levels.
Semantic Search
Semantic Search is a search method based on the meaning and intent of a query, rather than only finding documents whose keywords exactly match what the user typed.
Sentiment Analysis
Sentiment Analysis is a technique that helps AI identify the attitude or emotion expressed in text, typically classifying content as positive, negative, neutral, or into more specific emotional nuances.
Small Language Model (SLM)
Small Language Model (SLM) is a language model smaller in scale than an LLM, which typically requires fewer resources and lower operating costs. SLMs are well suited to small devices, internal systems, or specialized tasks.
Specialized Agent
Specialized Agent is an AI Agent designed for a specific domain or group of tasks, such as sales, customer care, or data analysis, with the appropriate knowledge, instructions, and tools.
Supervised Learning
Supervised Learning is a method of training AI with data that already has answers or category labels, helping the model learn to predict results for new data.
Supervisor Agent
Supervisor Agent is an AI Agent responsible for assigning tasks, tracking progress, and consolidating results from multiple other Agents to ensure the entire process achieves its goal.
Synthetic Data
Synthetic Data is data generated by models, algorithms, or simulated environments rather than collected directly from the real world. This data is commonly used to train and test AI.
System Prompt
System Prompt is a set of high-level instructions that defines the role, goals, limits, response style, and behavior that an AI model should prioritize following during a session.
T
2 terms
Tool Calling
Tool Calling is the capability that allows an AI model or AI Agent to select and request the use of external tools, such as search engines, calculators, databases, or APIs, to complete a task.
Transformer
Transformer is a way of structuring an AI model that helps the system identify relationships between parts of the data. This architecture is the foundation of many of today's language models and generative AI systems.
V
1 terms
Vision-Language Model (VLM)
Vision-Language Model (VLM) is an AI model that can understand images and text at the same time, enabling it to perform tasks such as describing images, answering questions, or reasoning based on visual content.
Z
1 terms
Zero-shot Learning
Zero-shot Learning is the ability of an AI model to perform a task or recognize a new category of data without being given training examples specifically for that task or category.
Get useful AI knowledge and trends for your business, straight to your inbox.
The latest AI trends and applications