Direct communication between autonomous agents to delegate work, share knowledge, and coordinate decisions without constant human involvement, enabling distributed, multi-agent workflows.
A predefined, automated action that fires when conditions are met, such as sending alerts, updating records, or kicking off workflows, so routine steps happen instantly.
Vector representations that encode an agent's role, skills, context, and history to enable smarter task routing, specialization, and matching across large agent pools.
The control layer that assigns tasks, manages agent interactions, prevents conflicts, and coordinates multi-agent execution so complex workflows run smoothly.
The reasoning module that decomposes goals into steps, chooses tools or agents, and sequences actions, adapting plans as context changes.
The capability that lets agents analyze problems, plan, adapt, learn from outcomes, and self-correct, turning them into proactive, goal-driven problem solvers.
End-to-end visibility into an agent's decisions, including model calls, tools, and context used, to support transparency, governance, and troubleshooting.
Marketing simple bots or scripted tools as "agentic AI" without real autonomy or reasoning, misleading buyers about actual capabilities.
AI systems that combine reasoning, planning, memory, and adaptability to pursue goals autonomously, breaking down tasks and acting across tools with minimal oversight.
Software powered by autonomous agents that make decisions and act in real time, often using LLMs, vision, or RL to handle complex tasks with limited human guidance.
Structured short- and long-term memory that lets agents retain context, recall past interactions and decisions, and personalize behavior over time.
A vendor-agnostic architecture where multiple autonomous agents collaborate across tools, systems, and models, making secure, real-time decisions at enterprise scale.
An approach that merges RAG with agent autonomy: agents decide what to retrieve, how to interpret it, and when to act, weaving retrieval, memory, reasoning, and decisions across multi-step tasks.
Goal-oriented task sequences planned and executed by agents that adapt on the fly to context and outcomes, needing minimal human input.
Applying agentic capabilities like planning and autonomy across domains so systems can manage complex tasks independently and coordinate actions without supervision.
Software entities that break down goals, use tools or APIs to execute steps, and deliver results with minimal human input, adapting to real-time data across systems.
Tools and dashboards that measure AI performance and behavior across interactions and outcomes, enabling tuning, A/B tests, and alignment with business goals.
A context-aware assistant that collaborates with users, offering suggestions, automating repetitive steps, and surfacing relevant insights to speed up work.
Practices that keep AI secure, ethical, and aligned with human values through robust design, governance, monitoring, and human oversight.
Training and testing AI in synthetic environments so agents can learn, explore edge cases, and improve safely without real-world risk.
High-performance infrastructure for training and running large models and generative workloads at scale for deep learning, reasoning, and real-time inference.
Gartner's framework for trust, risk, and security management across the AI lifecycle, covering governance, fairness, reliability, robustness, efficacy, and data protection.
An automated response to anomalies or thresholds, instantly notifying people or systems about issues like fraud, errors, or performance drops.
Attributing human traits like emotions or intent to machines, which can make interactions feel natural but also creates unrealistic expectations.
A standardized interface that lets your AI interact with software and data sources, enabling automated, cross-tool workflows.
A hypothetical AI that can understand, learn, and apply knowledge across tasks like a human, showing broad reasoning, creativity, and adaptability.
The field focused on systems that learn, reason, and make decisions, handling tasks without explicit rules for every scenario and improving over time.
Enriching models with external knowledge or tools at runtime for more accurate, grounded outputs without retraining.
A toolkit that automates common NLP tasks such as classification, sentiment, and intent detection with minimal setup.
A model that generates text token by token, conditioning each step on previous outputs; GPT-style models are the classic example.
Technology that converts spoken language into text in real time for voice assistants, IVR, and similar applications.
Agents that plan, act, and learn independently to achieve goals, adapting to changing conditions across multi-step processes.
Retrieval-Augmented Generation that improves answers by pulling relevant external info and using it to generate more accurate, grounded responses.
A standardized test to compare model performance on defined tasks and track progress objectively.
A keyword-based ranking algorithm that scores documents using term frequency and length, effective for classic exact-match retrieval.
Prompting models to reason step by step before answering, often improving accuracy and explainability.
Splitting large content into smaller, meaningful pieces to improve retrieval and RAG effectiveness.
A plug-in integration that links agents to third-party apps and data so they can take actions in real time without custom middleware.
Prebuilt AI capabilities like speech, vision, and translation that can be added to apps without training custom models.
Designing intents, prompts, and connectors as reusable modules to speed development and keep experiences consistent.
Building AI from modular components that can be rearranged and extended to scale and adapt quickly.
A measure of how certain the AI is about a prediction, used to choose next steps like proceeding, clarifying, or escalating.
Systematically capturing and applying user, task, and environment signals so agents act accurately and personally.
Logic that routes requests to the right agent, model, or workflow based on real-time context and confidence.
The amount of text a model can consider at once; larger windows support longer, more coherent reasoning.
Vector representations that reflect meaning based on surrounding context, improving retrieval and disambiguation.
The layer that aggregates roles, history, rules, and external data to inform agent decisions in the moment.
Methods to constrain and guide AI behavior to meet policy, compliance, and tone requirements.
Systems that engage via natural language across text or voice to understand intent and execute tasks.
Interfaces built around natural language interaction instead of menus or forms, common in chat and voice experiences.
Expanding training data by creating varied examples to improve robustness and performance.
Cleaning and structuring raw data so models can learn effectively.
Policies defining how long data is stored and when it's deleted for privacy and compliance.
A component that pauses to weigh alternatives before acting, improving decisions in complex workflows.
Retrieval using vector similarity to find semantically related content, not just exact keywords.
Releasing AI into production so real users can interact with it and performance can be observed.
A system that returns the same output for the same input, useful where consistency and auditability are crucial.
A visual tool to create conversational flows that connect user inputs to backend actions.
A guided conversation designed to complete a specific job, from gathering details to executing actions.
A model tailored to an industry or domain so it understands terminology and context for more accurate responses.
Running AI on local devices or servers close to the data for low latency, privacy, and offline resilience.
Models that produce embeddings, turning text and other data into vectors used for search, retrieval, and reasoning.
Vector encodings that capture meaning so systems can compare similarity, maintain context, and rank results.
Techniques that protect data by making it readable only to authorized parties during storage or transit.
AI designed for secure, reliable, large-scale use across business systems with governance and compliance.
RAG tailored for enterprises, grounding outputs in internal sources with security, traceability, and policy controls.
A structured piece of information, such as a name or date, extracted to make a request actionable.
Identifying and pulling key details from text to power routing and actions.
Building AI that is fair, privacy-respecting, accountable, and aligned with human values.
Techniques that make model decisions understandable and auditable, essential in regulated contexts.
A library of question-answer pairs used by assistants to deliver quick, accurate replies without full dialogues.
Training models across decentralized data on user devices or silos without moving the raw data, preserving privacy while aggregating learning.
Guiding a model to perform a new task using only a few in-prompt examples instead of retraining.
Adapting a general model to your data and requirements so it matches your tone, tasks, and domain.
Large, general-purpose models pre-trained on vast data that can be adapted to many downstream tasks.
The most advanced, often multimodal systems pushing state of the art in reasoning, planning, and autonomous behavior.
Systems that create new content text, images, code, and more by learning patterns from data.
A class of transformer-based language models pre-trained on large corpora and adaptable to many tasks.
RAG that leverages knowledge graphs to retrieve and reason over relationships, improving relevance and context.
Ensuring outputs are based on trusted sources and real context rather than unsupported generation.
Policies and controls that constrain model outputs and tool use to keep interactions safe, compliant, and on-brand.
When a model produces confident but incorrect or unsupported content; grounding and validation help prevent it.
Keeping people involved for review, approval, or intervention to maintain quality and compliance.
Combining keyword and semantic retrieval to capture exact matches and meaning-based results together.
Optimizing training settings like learning rate and model size to improve performance and efficiency.
A model learning a task from examples embedded in the prompt, without parameter updates.
Organizing content for fast retrieval by the AI during search and RAG.
Importing external data into the system so it can be indexed, searched, and used by agents.
Training models to follow natural language instructions more reliably and helpfully.
The user's underlying goal, used to select the right workflow or response.
Training related tasks together so they share representations and improve overall accuracy.
A centralized repository of vetted information that assistants use to answer questions consistently.
Structured networks of entities and relationships that help AI reason across connected facts.
Work focused on finding, understanding, and delivering information from the right sources.
Operational practices for deploying, governing, and improving language models across the enterprise.
A high-capacity model trained on large text datasets to understand and generate language across many tasks.
Managing how models interact with tools, memory, APIs, and agents to reason, retrieve, act, and adapt.
Persistent memory that lets agents remember preferences and past actions across sessions.
Visual tooling that lets non-developers build AI apps and automations with minimal coding.
A parameter-efficient fine-tuning technique that updates small matrices to customize large models cheaply.
Mechanisms for retaining and reusing information so agents stay context-aware over time.
Logic that picks the best model for a request based on task type, domain, or confidence.
Lifecycle management of models from training and deployment to monitoring and retirement.
Coordinating multiple specialized agents so they collaborate and hand off work effectively.
Collections of autonomous agents that communicate and share context to achieve broader goals.
Retrieval that uses multiple embeddings to capture different facets of meaning for better recall.
Systems that understand and generate across text, images, audio, or video for richer interactions.
Turning data or internal knowledge into fluent, human-readable text in real time.
Techniques that let machines parse, interpret, and work with human language.
Interpreting meaning and intent behind user input to drive accurate actions.
Building AI apps and workflows via drag-and-drop or forms without writing code.
Providing a continuous, consistent AI experience across chat, voice, web, email, and mobile.
A formal structure of concepts and relationships that helps systems reason within a domain.
Language models released for public use and customization, offering flexibility and control.
Learned values inside a model that determine how it represents language and generates outputs.
A model trained on large data that can be used as is or adapted to specific tasks.
A system that reasons in terms of likelihoods instead of single deterministic answers.
Linking prompts so the output of one step flows into the next to solve multi-step problems.
Crafting inputs that steer model behavior toward accurate, useful results.
Structured sequences of prompts and logic that complete end-to-end tasks reliably.
Refining inputs so retrieval is faster and more precise, reducing noise and ambiguity.
The ability to break down problems, follow steps, and make informed decisions under uncertainty.
Training via feedback signals for good or bad actions so agents improve through experience.
Combining RL with human preference signals to align outputs with expectations and values.
Practices ensuring AI is ethical, transparent, fair, and explainable.
A pipeline that retrieves trusted information, augments the prompt with it, and then generates grounded answers.
Rule-based automation that mimics user actions to perform repetitive tasks, without agentic reasoning.
Security that limits access to features and data based on user roles.
Structuring a complex task into smaller reasoning steps or helper modules so the model can solve it more reliably.
Retrieval that focuses on meaning and intent, not just exact keyword matches.
Detecting emotional tone in text to judge positivity, negativity, or neutrality.
Modeling ordered data to analyze or predict patterns where sequence matters.
Session-level context an agent uses during an active task to stay coherent.
Compact models optimized for specific tasks or constraints, offering speed, control, and privacy.
Tools, libraries, and docs that developers use to integrate and extend AI features.
Traditional keyword-based retrieval that excels at exact matches and speed.
Training with labeled examples so models learn to map inputs to correct outputs.
Creating artificial datasets for training or testing when real data is scarce or sensitive.
A generation setting that controls randomness: lower is more focused, higher is more varied.
Validating accuracy, safety, and edge cases before production use.
The text units models read and write; context and limits are measured in tokens.
A method where models learn when and how to call external tools by labeling tool-use in training data.
The examples a model learns from, which determine its knowledge and behavior.
Reusing knowledge from one task to accelerate learning on another.
A neural architecture using attention to model relationships in sequences, foundational to modern LLMs.
The ability to trace and explain how an AI arrived at its outputs to build trust and accountability.
Content like emails, PDFs, images, and audio that lacks a fixed schema, requiring AI to extract meaning.
Training without labels, where models discover patterns and clusters on their own.
Stores optimized for embedding vectors, enabling fast similarity search over meaning rather than keywords.
Finding relevant information by comparing embeddings to retrieve content by semantic similarity.
Handling tasks without prior examples by following instructions and generalizing from pretraining.