The AI Glossary: Essential Terms You Need to Know in 2026

AI is changing faster than plenty of us can keep up. Maybe you’ve nodded along in a meeting when someone mentioned transformers, embeddings, or fine-tuning, hoping context would fill in the gaps.

But if it didn’t, this glossary gives you a reliable reference for more than 30 essential AI terms organized by category, so the next time AI jargon comes up, you can follow the conversation and contribute to it. Bookmark it, share it with your team, and come back whenever you want a quick refresher. AI doesn’t need to feel like a riddle wrapped in a mystery.

Key takeaways

Here are the main points to remember:

• The AI glossary provides clear definitions of essential artificial intelligence terms, from foundational concepts like machine learning to emerging topics like agentic AI.
• Understanding the relationship between AI, machine learning, deep learning, and neural networks helps you navigate technical conversations with confidence.
• Practical terms like prompt engineering, hallucination, and temperature directly impact how you interact with AI tools daily.
• Domo’s AI capabilities, including DomoGPT and AI Chat, apply these concepts within a secure, governed data environment.

Why an AI glossary matters

AI isn’t just for experts. It’s for anyone curious about making more informed and timely decisions. Whether you’re sitting in a strategy meeting where someone drops ”retrieval-augmented generation (RAG)” into the conversation or trying to explain agentic AI to your team, having a reliable reference makes all the difference.

This guide, last updated in June 2026, organizes essential AI terms into five categories that range from foundational concepts to the Domo-specific tools that put these ideas into practice:

• Core AI concepts
• Language models and text generation
• AI accuracy and limitations
• Working with AI tools
• Domo AI terminology

Core AI concepts

These foundational terms establish the baseline for everything else in AI. Each definition follows a simple pattern: What the term means, where you’ll encounter it, and a concrete example to make it stick.

Artificial intelligence

Artificial intelligence (AI) refers broadly to technology that enables computers and machines to perform tasks typically requiring human intelligence. These tasks can include spotting patterns in data, making predictions, or even responding to your questions in real time. You can think of AI sort of as the brains behind the tools you use to help you work smarter and tackle challenges in new ways.

As you explore this glossary, you’ll also encounter related concepts like explainability and AI governance that shape how organizations deploy AI responsibly. It’s one of the key methods that helps AI systems learn from data.

Machine learning

Machine learning (ML) gives systems the ability to improve performance by learning from data. Imagine training a machine to predict customer trends by analyzing past behavior, improving its accuracy over time as it processes more information. It’s one of the key methods that make AI systems smarter.

You’ll see machine learning (ML) referenced whenever someone talks about training a model or improving predictions. A fraud detection system that gets more accurate at catching suspicious transactions the more examples it sees is machine learning in action.

Use ML when you have labeled data and a well-defined prediction task. Consider deep learning when your data is unstructured (images, audio, text) and you have sufficient volume. Assuming more data always means better results is a trap. Poor-quality or biased training data produces poor-quality predictions regardless of volume.

Deep learning

Deep learning uses neural networks with many layers to process complex patterns. It’s a subset of machine learning, but with a critical difference. Traditional ML might need humans to identify which features matter, like telling the system to look at transaction size for fraud detection. Deep learning figures out the important features on its own.

You’ll encounter deep learning behind image recognition, voice assistants, and language translation. When your phone unlocks by recognizing your face, that’s deep learning analyzing thousands of facial features simultaneously.

Neural networks

Neural networks are computing systems loosely inspired by the human brain. Made up of interconnected nodes (neurons) organized in layers, information flows through these layers, with each node performing simple calculations and passing results forward.

Neural networks are the underlying architecture for most modern AI. When a streaming service recommends shows based on your viewing history, neural networks are processing those patterns across millions of people to find what you might like next.

Natural language processing

Natural language processing (NLP) enables computers to understand, interpret, and generate human language. It bridges the gap between how humans communicate and how machines process information.

You’ll encounter NLP in chatbots, email filters, sentiment analysis tools, and voice assistants. When your email automatically suggests replies or flags a message as spam based on its content, NLP is doing the heavy lifting.

Language models and text generation

Modern large language model (LLM) applications are built from several interconnected components. Tokens feed into context windows, embeddings power retrieval, and transformers make it all possible. These terms work together as a system.

Large language model

Large language models are advanced AI systems trained on extensive sets of data, such as text from books, studies, online articles, transcriptions and more. Because they learn from such massive troves of text data, these models discern the patterns and nuances in grammar and language.

This allows them to perform complex tasks like summarizing a transcript from a call or translating an email into a different language. Chatbots, for example, can use LLMs to generate natural-sounding responses to questions, providing a human-like conversational experience.

When you’re choosing an LLM for a specific task, understanding how tokens and context windows work (covered below) helps you pick the right model for your needs.

Generative AI

Generative AI produces something entirely new, whether that’s a text-based report, an image, or even a piece of music, based on what it’s learned from data. Popular tools like ChatGPT show how generative AI can create text and images, while Claude demonstrates how AI can build workflows and tools. But Domo adds governance controls for business data.

You’ll encounter generative AI whenever you’re asking a tool to produce content rather than just analyze it. Its ability to move beyond analysis and actually produce original content can help with everything from creative brainstorming to drafting emails. Don’t confuse generation with accuracy, though. Generative AI excels at producing plausible content, not necessarily factual content.

Tokens

LLMs don’t read words the way humans do. They break text into smaller chunks called tokens, which might be whole words, parts of words, or even punctuation marks.

You’ll encounter tokens when you hit usage limits or see pricing based on token counts. The sentence ”AI is powerful” contains approximately four tokens. Understanding tokens helps explain why longer prompts cost more and why there are limits to how much context a model can handle at once.

Context window

The context window represents the maximum amount of text an LLM can consider at one time, measured in tokens. Think of it as the model’s working memory: Everything it needs to reference must fit within this window.

You’ll see context window limits when working with long documents or complex conversations. When a document exceeds the context window, chunking strategies determine what the model actually sees, which directly affects output quality. These limits are why retrieval-augmented generation matters for enterprise applications.

Embeddings

Numerical representations of text (or other data), embeddings allow models to understand semantic similarity. When you ask an AI a question, your query is converted to an embedding and matched against a database of similarly embedded documents.

You’ll encounter embeddings behind the scenes in search, recommendations, and RAG systems. Two sentences with similar meaning will have embeddings that are mathematically close, even if they use different words.

Transformer model

Transformer models are the architecture powering modern LLMs, introduced in 2017 and now the foundation for Generative Pre-trained Transformer (GPT) models, Claude, and most other language models. Their key innovation is the attention mechanism, which allows the model to weigh the importance of different words in relation to each other.

You’ll see transformers referenced in technical discussions about model architecture. ”Transformer-based” describes this specific approach to processing language that revolutionized what AI could do with text.

AI accuracy and limitations

Before deploying AI in any serious capacity, you should understand where it can go wrong. These terms form the ”know before you deploy” layer, connecting directly to the mitigation strategies covered in the RAG and fine-tuning sections.

Common AI limitation categories include:

• Output accuracy (hallucination, factual errors)
• Fairness and representation (bias, skewed training data)
• Transparency (explainability, auditability)
• Reliability (consistency, edge case handling)

Hallucination

Hallucination happens when AI generates false or fabricated information presented as fact. Language models predict plausible-sounding words based on patterns they've seen. They don’t understand truth or have access to verified information.

This means you’ll encounter hallucinations when AI confidently cites sources that don’t exist or invents statistics that sound reasonable but are completely made up. A model asked about a company’s quarterly earnings might generate a believable number that has no basis in reality. The confidence with which these models present fabrications makes them particularly dangerous in high-stakes contexts, a risk that many guides overlook.

To reduce hallucination, verify outputs with RAG-based retrieval, apply grounding to anchor responses to source documents, and use citation requirements so the model must reference its sources.

Grounding

Grounding anchors AI outputs to verified, specific source material, reducing the likelihood of hallucination. A grounded response traces back to a real document, dataset, or knowledge base.

You might see grounding referenced in enterprise AI deployments where accuracy matters. A grounded answer to ”What was our deal status?” cites the specific sales call report it pulled from, rather than generating a plausible-sounding number with no citation. Grounding represents the output quality standard; RAG is one of the retrieval mechanisms that helps achieve it.

Bias in AI

Bias in AI occurs when systems produce skewed or unfair outputs, typically because of imbalances in training data or problematic patterns the model has learned to replicate.

“Training data bias“ means skewed input data produces skewed outputs. For example, if a hiring model is trained primarily on resumes from one demographic, it may unfairly favor that group. “Output bias“ means the model amplifies patterns in ways that disadvantage certain groups, even if the training data seemed balanced.

Mitigation involves both data governance practices (diverse, representative training data) and model evaluation (testing outputs across demographic groups before deployment). Teams often test for bias once at launch and assume the problem is solved. Bias can emerge over time as models encounter new data patterns.

Explainability (XAI)

Explainability, also called XAI or explainable AI, describes the ability to articulate how and why an AI system reached a particular output in terms a human can understand. It’s distinct from transparency (knowing what data was used) and interpretability (understanding the model’s internal mechanics).

You may notice explainability requirements in regulated industries where AI-driven decisions, like credit approvals or medical recommendations, must be auditable. This means an explainable AI system for a bank, for example, should both flag a loan application as high-risk and tell you which factors drove that score.

Working with AI tools

Three approaches dominate how people customize AI behavior, and they’re not mutually exclusive.

• Prompt engineering requires no training data and works immediately.
• Retrieval-augmented generation (RAG) adds real-time retrieval without retraining the model.
• Fine-tuning requires labeled data and compute but produces the most consistent domain-specific behavior.

These terms matter for anyone actively using AI systems.

Prompt engineering

Prompt engineering is the practice of crafting inputs to get stronger outputs from AI systems. Small changes in phrasing can dramatically affect quality and relevance. So does your level of specificity. Being too vague kills quality.

You use prompt engineering every time you interact with a chatbot or AI assistant. Instead of asking ”Write me an email,” a well-engineered prompt might be: ”Write a professional but friendly email to our client, Sarah Chen at Acme Corp, explaining a two-week delay in the Q2 dashboard rollout due to data integration challenges. Emphasize our commitment to quality and accuracy, propose a revised timeline of March 15, and offer a brief progress update call this week. Keep the tone apologetic but confident, around 150 words.”

Fine-tuning

Fine-tuning takes a pre-trained AI model and trains it further on your specific data to improve performance for particular tasks. You might fine-tune a model when off-the-shelf models don’t quite match your needs. Choose fine-tuning when you want consistent tone or domain vocabulary and have labeled training data. RAG is a good fit when your data changes frequently or you need source attribution.

Temperature

Temperature controls how creative or random an AI’s outputs are. Lower temperatures (closer to 0) produce more predictable, focused responses. This is ideal for factual Q&A, to keep answers consistent. Higher temperatures produce more varied, creative outputs, ideal for brainstorming or creative writing. You’ll adjust the temperature when you want different types of responses from the same model.

Retrieval-augmented generation

Retreival-augmented generation (RAG) combines the creative power of AI with real-time data retrieval. This AI framework improves the answers generated by a large language model (LLM) by providing it with outside sources of data that may be more relevant. Think of it as pairing an AI’s ability to generate content with live, accurate information.

For instance, RAG can pull the latest sales data while helping craft a business summary, keeping the output current and reliable. A customer service bot using RAG can answer questions about your specific products by retrieving information from your knowledge base rather than relying on what the model learned during training. When RAG fails, it’s usually because of poor chunking or missing reranking.

Vibe coding

Vibe coding describes the practice of using AI tools to generate code through conversational prompts rather than writing syntax manually. Instead of memorizing programming languages, you describe what you want the code to do, and the AI produces working code based on your intent. You might be vibe coding when using AI assistants to build scripts, automate workflows, or prototype applications using just your natural language.

For example, you might tell an AI ”Create a Python script that pulls sales data from our API and exports it to a CSV file,” and receive functional code back. This approach lowers the barrier to automation, letting business users solve technical problems without becoming developers. However, vibe coding is limited because it doesn’t natively or securely connect to your actual business data.

Model context protocol

Model context protocol (MCP) is an open standard that enables AI systems to securely connect to external data sources and tools in a consistent way. Instead of building custom integrations for each data source, MCP provides a universal framework for AI models to access the information they need.

MCP is what connects AI agents to business systems like databases, APIs, or file repositories. For example, an AI agent using MCP could pull customer data from your CRM, sales figures from your data warehouse, and product information from your content management system, all through a single standardized protocol. This makes it easier to build AI workflows that span multiple data sources without creating security vulnerabilities or maintenance headaches.

Agentic AI and automation

Beyond creating information, AI can also power your processes and solve your workflow problems. Agentic automation systems often require human-in-the-loop checkpoints to maintain oversight, especially in enterprise settings where decisions have significant consequences.

Agentic AI

Agentic AI systems understand your goals, reason through which actions should be taken to achieve your objectives, and execute on those decisions. Because these systems can operate independently, agentic AI helps you work more efficiently and frees up your time to devote to more important tasks.

Consider a workflow where you need to analyze customer feedback, identify trends, and create a summary report. An agentic AI system could handle the entire sequence: pulling data, running analysis, drafting the report, and flagging it for your review before sending.

AI agents

If ”agentic AI” describes the concept of powering your workflows with AI, AI agents are the tools that make this idea a reality. They‘re the specialized systems that perform the actual tasks like problem-solving and execution.

AI agents use LLMs to understand and respond to what you might ask of them, and they also know when to use external tools to achieve their goals. Another benefit of AI agents is that they learn from interactions, so they improve over time at meeting your expectations.

AI agent capabilities typically include:

• Understanding natural language requests.
• Breaking complex goals into actionable steps.
• Accessing external tools and data sources.
• Learning from feedback to improve performance.

Multi-agent systems in AI

In a multi-agent system, different AI agents work together to achieve more complex goals. Each agent has a unique role. One might analyze customer trends while another forecasts inventory requirements. Together, they share resources and learn from each other, which allows them to take on more challenging tasks and perform more effectively than they would individually.

AI agent marketplace

An AI agent marketplace functions like an app store for automation. Instead of coding your own solutions, you can browse, select, and deploy pre-built AI agents designed to handle specific workflows.

Multimodal AI

Multimodal AI refers to systems that can process and generate multiple types of content, text, images, audio, and video, within a single model. Rather than needing separate tools for each format, multimodal models handle them together.

For example, multimodal AI tools let you upload an image and ask questions about it, or generate images from text descriptions. A model that can analyze a chart, read the text in it, and answer questions about the data is demonstrating multimodal capabilities.

Reinforcement learning

Reinforcement learning is a type of machine learning where an AI learns by taking actions and receiving feedback in the form of rewards or penalties. Unlike supervised learning, which requires labeled examples, reinforcement learning discovers optimal behavior through trial and error.

Reinforcement learning is behind gameplaying AI, robotics, and recommendation systems that adapt based on user behavior. When a model learns to play chess by playing millions of games against itself and improving based on wins and losses, that’s reinforcement learning.

Domo AI terminology

AI might feel overwhelming at first, but it’s really just a set of tools designed to help you get more done with less effort. Whether you’re looking to automate repetitive tasks, improve decision-making, or uncover insights, AI can make it happen.

Domo’s AI capabilities are designed to be approachable and practical, so you can focus on the outcomes more than the learning curve.

Also, note that Domo’s AI ecosystem connects directly to the governance vocabulary covered earlier in this glossary: auditability, access control, and data lineage are built into how these tools operate.

Domo AI

You can think of Domo AI as an umbrella term for all of Domo’s AI capabilities. It encompasses everything from the AI Service Layer to individual agents, unified under a secure and governance-first approach.

AI governance

AI governance refers to the policies, processes, and controls that ensure AI systems are used responsibly, ethically, and in compliance with organizational and regulatory requirements. It covers everything from data access permissions to audit trails to bias monitoring.

Domo operationalizes AI governance through built-in access controls, audit logging, and data lineage tracking. Rather than bolting governance on after the fact, these controls are embedded in how Domo AI functions, so you can deploy AI capabilities with confidence that oversight is already in place.

AI Service Layer

Domo’s AI Service Layer is a flexible AI framework built to simplify and streamline data exploration in your dashboards and apps. The AI Service Layer provides access to a variety of services, such as text generation and summarization, and gives you a space to create and train your own AI models.

Available services include natural language querying, automated insights generation, and custom model deployment, all within Domo’s governed environment.

AI Activation Layer

The AI Activation Layer is the suite of tools inside Domo that help you use AI agents to activate your data foundation. It includes Analyzer, App Studio, Workflows, and more. This layer is how AI can interact with your data in a secure and governed way. All of it happens on top of your cloud data platform.

Semantic Layer

The Semantic Layer is the foundation of usable, trustable AI. In Domo, it’s a middle layer that sits in between your data and your AI use cases. The Semantic Layer contextualizes and grounds your data in reality by mapping raw technical data to familiar business terms. This is so that your metrics, KPIs, and calculated fields stay consistent across your enterprise.

DomoGPT

ChatGPT can generate natural language responses quickly, but DomoGPT keeps business data inside the Domo environment with private model routing and access controls.

To ensure that data from people using Domo remains secure, Domo built DomoGPT, which uses a suite of DomoCloud private models to keep your data within the Domo ecosystem, protected from third-party access.

DomoGPT addresses enterprise governance requirements through private model routing to prevent data exfiltration, aligning access controls with your existing permissions, and providing audit logging for compliance. Your sensitive data never leaves the Domo environment. Many organizations learn this lesson the hard way with other tools.

Conversational Agents

Domo’s Conversational Agents are chat-based AI agents that pop up in the contexts where you work. Each one is a master of a single business domain, and it speaks that domain’s language and understands the data. You don’t have to give it additional context. So, for example, you can have a conversational agent trained on sales pipeline analysis and ask it questions while you’re reviewing pipeline data in Domo. It’ll give you clear, contextualized answers back with fewer hallucinations.

Conversational Agents are deployed in Domo’s AI Library, a central, governed environment for creating and managing them. From there, you can deploy them anywhere they’re needed.

AI Data Dictionary

Prepare your datasets so that they have all of the information necessary for AI tools to draw insights from them. This AI-readiness feature helps add metadata, descriptions, and other data dictionary items that allow the AI Chat agent to return more accurate answers.

The AI Data Dictionary functions as a governed data catalog layer that supports lineage tracking and AI-readiness auditing. By documenting what your data means and where it comes from, you enable more accurate AI responses and maintain the governance standards enterprise deployments require.

Putting AI knowledge into action

AI might feel overwhelming at first. Really, it’s just a set of tools designed to help you get more done with less effort. Whether you’re looking to automate repetitive tasks, improve decision-making, or uncover insights, AI can make it happen.

And you don’t need to figure it out alone. Domo’s AI capabilities are designed to be approachable and practical, so you can focus on the outcomes, not the learning curve.

Now that you’ve got the basics, take the next step. Get a demo to explore how Domo can help you turn these ideas into action.