The AI industry moves fast, and so does the terms used to describe it. This glossary helps you stay up-to-date with the evolving market language.
How AI agents communicate and hand off tasks to each other without human involvement.
How often an AI system produces the correct output for a given task.
A predefined, trigger-based AI task that runs automatically for routine actions without any manual effort.
A retrieval approach that adjusts its strategy based on the complexity of each query.
Vector-based representations of an agent's role and skills, used to route tasks to the most suitable agent.
AI systems that can plan, reason, and act autonomously to complete multi-step goals.
AI-powered software where autonomous agents take actions and make decisions in real time.
An agent's ability to store and recall context within a session or across multiple sessions over time.
A flexible, vendor-agnostic AI architecture where multiple autonomous agents collaborate across tools and systems.
The coordination of multiple AI agents, tools, and humans to complete a larger goal.
A retrieval method where agents actively decide what to retrieve, when, and how.
How an AI agent thinks through a problem, breaks it into steps, and adjusts its approach as it goes.
Multi-step processes run and adapted by AI agents with minimal human direction.
Bringing reasoning, planning, and autonomy into any specific application or business domain.
The full journey of an AI agent from creation and testing through deployment and retirement.
The tools and processes used to monitor, update, and govern deployed AI agents.
A controller that assigns tasks to the right agents and coordinates how they work together.
The component of an agentic system that turns a high-level goal into a step-by-step action plan.
How an AI agent analyzes a problem and decides on the best course of action.
The ability to track and audit every decision and action taken by an AI agent.
Marketing simple bots or rule-based tools as agentic AI when they don't meet that standard.
Software that can independently execute tasks using tools, APIs, and reasoning — without step-by-step human direction.
Tools and dashboards that monitor AI performance across interactions, intent detection, and resolution outcomes.
Enriching AI models with external context, real-time data, or tools to produce more accurate and relevant outputs.
An AI assistant that works alongside a human to surface insights, draft content, and reduce manual effort.
The policies and controls that ensure AI systems operate safely, fairly, and within regulations.
When an AI generates a confident but incorrect or fabricated output.
The practice of building AI systems that behave reliably and avoid causing unintended harm.
Using synthetic environments to train and test AI models safely before real-world deployment.
High-performance infrastructure built to train and run large AI models at scale.
A framework for managing AI trust, risk, and security across the full AI lifecycle.
An AI-triggered response to anomalies or threshold breaches that notifies the right people or systems automatically.
Attributing human traits like emotions or consciousness to AI systems.
A standardized interface that lets two software systems exchange data and trigger actions.
Checkpoints in an automated process that require a human to review and authorize before the workflow continues.
A theoretical AI capable of understanding and applying knowledge across any task — the way a human can.
Technology that enables machines to perform tasks that typically require human reasoning and judgment.
A time-stamped log of every action and decision made by an AI system or user.
Technology that converts spoken language into text in real time.
Using software to handle tasks and processes that would otherwise require human effort.
A toolkit that automates key NLP tasks like classification and intent detection with minimal setup.
A model that predicts the next value in a sequence by learning from previous values — the basis of most modern language models.
Automated processes for tasks like loan processing, KYC checks, account servicing, and fraud detection.
A retrieval approach that enhances language model responses by pulling relevant information from external sources before generating an answer.
A standardized test used to measure and compare how well AI models perform on specific tasks.
A keyword-matching retrieval algorithm often used alongside vector search for more complete results.
A technique that instructs an AI to reason step-by-step before producing a final answer.
Breaking large documents into smaller pieces so they can be indexed and retrieved effectively by AI systems.
Categorizing input data into predefined labels — a foundational task in AI-powered routing and automation.
A plug-and-play integration that links AI systems to third-party cloud apps and databases.
Pre-built AI capabilities for tasks like speech recognition, image analysis, and language translation.
Ensuring AI-driven processes operate within legal, regulatory, and internal policy requirements.
Automated sequences that enforce regulatory rules, capture required approvals, and maintain documentation.
Building AI elements that can be reused across agents or applications to speed up development and ensure consistency.
Building AI capabilities as modular, reusable blocks that can be assembled and scaled for different use cases.
A numerical indicator of how certain an AI system is about a particular output.
Designing systems to give AI agents the right background information at the right time.
A system that directs requests to the right agent, model, or workflow based on contextual signals.
Transforming words or data into numerical vectors that capture meaning based on surrounding context.
A system that collects and analyzes signals to give AI agents memory, situational awareness, and smarter decision-making.
The maximum amount of text an AI model can read and process in a single interaction.
How well AI behavior can be guided or constrained through defined boundaries.
Technology that enables machines to interact with humans using natural language across text, voice, and messaging.
A user interface that enables interaction through natural language instead of buttons or forms.
Using AI agents to handle support tickets, FAQs, and service requests without requiring human agents for every interaction.
Expanding training data by generating or modifying existing examples to improve model robustness.
Cleaning and formatting raw data so AI models receive structured, consistent inputs.
Controls that govern how sensitive data is stored, accessed, and used by AI systems.
Defining how long user and system data is stored before it is deleted.
AI-driven logic that makes routine decisions automatically based on defined rules or model output.
A system that enables AI agents to evaluate options before acting on complex or high-stakes workflows.
Using vector embeddings to find semantically similar information rather than matching keywords.
When an AI system goes live for real users in a production environment.
The distinction between rule-based outputs that are fixed and model-generated outputs that can vary.
A visual workspace for designing conversation flows using drag-and-drop tools without writing code.
A guided conversation path designed to complete a specific user goal through defined logical steps.
AI systems trained or configured to work within a specific industry or business function.
A language model fine-tuned for a specific industry to deliver more accurate, relevant responses.
Running AI directly on local devices rather than in the cloud, enabling fast and private real-time decisions.
Models that convert language into numerical vectors, allowing AI to find connections based on meaning.
Numerical representations of text or data that capture semantic meaning — the foundation of modern AI search.
Scrambling data so only authorized parties with the correct key can access it.
AI technology designed to meet the scale, security, and compliance requirements of large organizations.
Combining intelligent retrieval with LLM-generated responses from internal, organization-specific knowledge bases.
A search system that lets employees find information across internal data sources using natural language.
A specific piece of information the AI extracts from user input, such as a name, date, or account number.
Identifying and pulling key details from user inputs to help AI accurately route and process tasks.
Building AI systems that are fair, transparent, responsible, and aligned with human values.
Testing AI outputs against quality, accuracy, and reliability benchmarks — before and after deployment.
AI that can show why a decision was made, rather than leaving the reasoning opaque.
Pulling structured data from unstructured sources like documents, emails, or forms.
Pre-trained question-answer pairs used by virtual assistants to deliver fast, accurate responses to common queries.
Training AI models across distributed data sources without centralizing sensitive data in one place.
Enabling AI models to understand new tasks from just a handful of examples provided in the prompt.
Training a pre-built AI model further on domain-specific data to improve its performance for a particular use case.
Large, general-purpose AI models trained on massive datasets and adaptable to many different tasks.
The most advanced AI systems available, pushing the boundaries of reasoning and autonomous capability.
Enabling an AI model to trigger specific tools or APIs when it needs real-world data or needs to take action.
AI systems that create content — text, images, code, or audio — by learning patterns from existing data.
A family of generative language models capable of understanding and producing human-like text across many tasks.
A reasoning framework that maps thinking as a graph, allowing multiple interconnected paths instead of a single linear chain.
Combining retrieval-augmented generation with knowledge graphs to improve reasoning across connected information.
Ensuring AI outputs are based on verified, trusted sources rather than model-generated assumptions.
Rules and filters that constrain AI output to stay within safe, accurate, and compliant boundaries.
Automating the steps involved in bringing a new employee into an organization — from document collection to system access.
A design pattern where humans review or approve AI outputs at defined points in a workflow.
Combining keyword-based and semantic search to retrieve both exact matches and meaning-based results.
Optimizing the settings that control how an AI model learns, to improve accuracy and performance.
AI-driven handling of IT requests, password resets, access issues, and common troubleshooting tasks.
Automated capture, validation, and routing of invoices through the finance approval workflow.
What a user wants to accomplish — the signal AI uses to determine the right response or workflow.
Connections between an AI platform and external systems like CRMs, ERPs, and databases.
Using AI to process, validate, and assess insurance claims with less manual effort and faster turnaround.
Training AI models to follow human instructions more effectively and produce consistently useful outputs.
Importing external documents and data into an AI system to make that content searchable and usable.
Pulling specific facts or data points from unstructured content like documents or emails.
Organizing and storing data so AI can quickly search and retrieve it at query time.
Enabling AI models to handle new tasks by reading examples within the prompt — no retraining required.
Training multiple AI tasks or models simultaneously by sharing knowledge across them for better overall performance.
When AI finds, understands, and delivers information by connecting to the right sources.
Structured frameworks that show how concepts, entities, and data points relate to each other.
A structured repository of information that AI agents can search to answer questions accurately.
A technique for fine-tuning large models efficiently without retraining the entire model from scratch.
Platforms that let users build AI-powered applications using visual interfaces with minimal coding required.
An agent's ability to remember information and context across multiple interactions and sessions.
Practices and tools used to deploy, manage, and monitor large language models in production.
Managing how large language models interact with tools, memory, APIs, and agents across a workflow.
Automatically assigning inbound leads to the right sales rep based on defined criteria and context.
The time delay between when a request is submitted and when the AI responds.
An advanced AI trained on massive text datasets to understand, process, and generate human language.
The practice of managing and optimizing how language models are deployed and used across the enterprise.
Improving retrieval accuracy by using more than one semantic representation to find relevant information.
AI systems that can understand and process more than one type of input — like text, images, and audio together.
Architectures where multiple specialized AI agents collaborate to complete a larger, more complex task.
Ongoing observation of AI system performance, accuracy, and behavior once it's in production.
Managing the full lifecycle of AI models — from training and testing to deployment and retirement.
Choosing the right AI model for a task based on capability, cost, and latency requirements.
A system that decides which AI model to use for a specific task based on prompt type, complexity, or cost.
An agent's ability to retain and reuse information from earlier in a conversation or from past sessions.
An open standard that enables AI applications to connect with external data sources and tools through a unified interface.
AI-assisted monitoring, scheduling, and exception handling across production and factory floor operations.
Platforms that let users build AI applications or automations using visual tools — no coding knowledge needed.
A subset of NLP focused on interpreting the meaning and intent behind what someone says or writes.
The field of AI focused on understanding, interpreting, and generating human language.
Turning structured data or internal knowledge into clear, human-readable language.
Coordinating agents, tools, models, and data sources to complete multi-step workflows as a unified system.
Large language models freely available for anyone to use, customize, or self-host.
A structured framework for organizing knowledge and defining relationships among concepts in a specific domain.
Running AI systems within a company's own infrastructure rather than a shared public cloud.
Delivering a consistent AI experience across chat, voice, email, web, and mobile — with context maintained throughout.
The ability to see what an AI system is doing, why it made certain decisions, and how well it is performing.
Structured sequences of prompts, logic, and decision steps that together drive a larger automated task.
The craft of writing and structuring prompts to get more reliable and useful AI outputs.
Linking multiple prompts together where the output of one becomes the input to the next.
The input instruction given to an AI model to guide what it produces.
AI-driven handling of purchase requests, vendor comparisons, approvals, and purchase order generation.
A model that makes decisions based on the likelihood of different outcomes rather than fixed rules.
An AI system already trained on large datasets that can be used directly or fine-tuned for specific tasks.
Internal values a language model learns during training that control how it interprets inputs and generates outputs.
Refining a query before it's processed so the AI retrieves more precise and relevant results.
The rules or model-driven criteria that determine where a request is sent within a system.
Directing a request to the right agent, model, or team based on its content or context.
Restricting what different users or systems can see or do within an AI platform based on their role.
Automating repetitive, rule-based tasks using bots that mimic human actions in software systems.
Pre-built agents, prompts, or workflow blocks that can be deployed across multiple use cases without rebuilding.
Finding the most relevant documents or data in response to a query — the first step in most RAG systems.
Building and deploying AI systems that are ethical, transparent, fair, and designed with human impact in mind.
Improving AI behavior by training it on how humans rate and correct its outputs.
A training method where AI learns by trial and error, improving based on rewards and penalties over time.
The process by which an AI agent analyzes a problem and decides on the best course of action.
Combining a language model with a retrieval system so it can answer using up-to-date or organization-specific data.
Creating artificial data to train or test AI models when real data is limited, sensitive, or hard to obtain.
Real-time awareness of where goods, components, or orders are at any point in the supply chain.
Training an AI model using labeled examples where both the input and correct output are known.
The distinction between data in defined formats like tables and free-form content like emails or documents.
Keyword-based retrieval that's fast and effective for exact matches like specific codes or product names.
A collection of tools, libraries, and documentation that helps developers build or extend AI applications.
Compact AI models built for specific tasks — faster, cheaper, and easier to control than large general-purpose models.
Storing recent inputs and conversational context that an agent uses during an active session.
Analyzing or predicting patterns in ordered data where the sequence itself affects the outcome.
Helping AI understand the emotional tone behind text — whether it's positive, negative, or neutral.
Finding content based on meaning and intent rather than exact keyword matches.
An approach where an AI system evaluates and refines its own retrieval and generation process before responding.
Breaking a complex task into smaller steps that the AI can reason through using intermediate prompts.
The ability of an AI system to handle growing volumes of work without degrading in performance.
Using AI to evaluate whether an inbound lead meets the criteria to become a real sales opportunity.
A reasoning approach where AI explores multiple solution paths before selecting the best outcome.
The ability to see and understand how an AI system arrived at its outputs.
The model architecture behind most modern large language models, enabling AI to understand how words relate to each other in context.
When an AI applies knowledge learned from one task to perform better on a different one.
The data an AI system learns from to understand language and produce accurate outputs.
When AI generates harmful, offensive, or inappropriate content.
An AI model's ability to interact with external tools, APIs, or functions to complete tasks beyond text generation.
The ability of an AI model to invoke external tools, APIs, or systems to retrieve data or take action.
The chunks of text that an AI model reads and writes — roughly equivalent to words or parts of words.
Verifying that an AI system works as expected before it goes live by checking accuracy, behavior, and edge cases.
A setting that controls how predictable or varied an AI model's responses are.
Assigning a subtask to another agent or tool within a larger agentic workflow.
Training AI on unlabeled data so it learns by spotting patterns and relationships on its own.
Information without a fixed format — like emails, PDFs, or audio files — that AI makes usable at scale.
Finding information based on meaning by comparing numerical representations of content.
A database that stores and searches data as embeddings rather than traditional rows and columns.
Automating a sequence of steps across systems so they execute consistently without manual intervention.
Letting an AI handle tasks it hasn't been explicitly trained on by relying on instructions alone.
