New AI labels arrive faster than most people can learn them, and nobody has measured exactly how fast. The feeling of falling behind is reasonable, but the fix isn’t memorizing more terms. Most of the vocabulary comes down to a few distinctions: how broad a term is, what the system produces, whether it can act, and who is defining it. This article covers those four and gives you a short check to run on any unfamiliar term.
The map: from AI down to the LLM
Think of nested circles, from broadest to narrowest.
As an Amazon Associate I earn from qualifying purchases.
Artificial intelligence (AI)
This is the umbrella. Even here there is no single wording. NIST’s glossary search returns several definitions of “AI” from different sources, because legal, technical and policy documents each frame it for their own purpose. Treat any one-line definition as a convenient summary, not the final word.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsGenerative AI (GenAI)
Generative AI is the part of AI concerned with creating new content. Google Cloud’s generative AI glossary describes it as using foundation models to produce new text, images, audio or video. NIST’s GenAI glossary entry points to its publication NIST AI 100-2e2025 as the source.
#1 Best Overall
Large language model (LLM)
An LLM is a language-focused kind of generative AI. The UK Information Commissioner’s Office (ICO) glossary describes LLMs as a type of generative AI that can produce human-like text, code and translations. The common mistake is treating “LLM” and “generative AI” as synonyms. An image generator is generative AI but not an LLM.
Foundation model
This term describes how a model is built and reused, not what it outputs. NIST defines a foundation model as one trained on broad data using self-supervised learning, which can be adapted to many downstream tasks, including through fine-tuning. The NIST entry is tied to NIST AI 100-2e2025. Many LLMs are foundation models, and Google Cloud’s glossary treats foundation models as the basis for generating new content. Still, “foundation model” answers “how general and adaptable is it?” while “LLM” answers “what kind of content does it handle?”
Rank #2
Model versus agent: the layer that acts
A model produces output when given input. An AI agent is a larger thing built around a model. Google Cloud describes an agent as an application that processes input, reasons using available tools, and takes actions based on its decisions to achieve a goal. It breaks agents into components such as orchestration, a model, and tools.
That gives you a practical test: does the product only answer, or can it use tools and carry out steps toward a goal? Note that this is a vendor’s explanation. It’s clear and useful, but it doesn’t establish an industry standard.
Why “agentic AI” means different things
“Agentic AI” is a growing label for systems with some degree of goal-directed action or autonomy, and it has no universally settled definition. The OECD’s working paper The agentic AI landscape and its conceptual foundations (13 February 2026) surveys how existing sources define “agentic AI” and “AI agent.” It compares the definitions and identifies recurring features rather than treating them as identical.
So don’t assume that two products both called “agentic” share the same autonomy. One may draft an email for your approval. Another may send it and update your calendar unprompted. Ask what the system can actually do, and how much a human checks.
Comparing terms on five axes
| Axis | Question to ask | Examples |
|---|---|---|
| Scope | Is it a field, a model type, or an application? | AI (field); foundation model (model type); agent (application) |
| Output | What does it produce? | LLMs: text, code, translations; generative AI overall: text, images, audio, video |
| Adaptability | Can a broadly trained model be adapted to specific tasks? | Foundation models, including via fine-tuning |
| Action | Does it only respond, or use tools and act toward a goal? | Agents and “agentic” systems |
| Authority | Who is defining the term? | NIST, OECD, ICO, Google Cloud, Brookings |
Check who is defining the term
Definitions carry different weight depending on where they come from:
Recommended Free Tools
- NIST glossary entries point to a named source document (NIST AI 100-2e2025), so you can trace them.
- The OECD paper is a policy analysis that deliberately compares definitions, which makes it useful for contested terms.
- The ICO glossary is a regulator’s plain-language reference. No publication or revision date was established for it.
- Google Cloud documentation is vendor material, updated continuously. It is good for accessible explanations, but it reflects one company’s framing.
- Brookings publishes a glossary with additions dated 30 August 2024, 15 March 2025 and 20 January 2026. It states that its definitions are not official or authoritative.
A three-question check for any new buzzword
This is a working method drawn from the differences above, not an official standard. You can set a new label aside until you can answer these:
Best Value
- What capability does it describe? Generating text, adapting to tasks, using tools, acting on its own?
- What does it refer to? A field, a model, or a finished application built on one?
- Whose definition is this? A standards body or regulator, a vendor, or marketing shorthand? Check the source and its date.
If a speaker can’t answer the first question in plain words, the term is probably doing more selling than explaining. Most new labels turn out to be a combination of the distinctions you already have.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
