AI glossary
Plain-English definitions of the AI terms you meet at work, written by Ronan Liedmeier for AI Academy. Every entry starts with a one-sentence definition and links to the module that teaches the idea in depth.
By Ronan Liedmeier, fractional head of growth and course author · Last updated 17 October 2026
A
- Agent harness
- An agent harness is the software wrapped around an AI model that turns it into a working agent: it supplies the tools, file access, permissions, memory and the loop that chains actions together. The model does the reasoning; the harness decides what it can see and do.
- Agent loop
- An agent loop is the repeating cycle an AI agent runs to finish a task: it plans a step, calls a tool, reads the result and decides whether to continue or stop. The loop ends when the goal is met or a limit is reached.
- Agent skill (SKILL.md)
- An agent skill is a folder of instructions, built around a SKILL.md file, that teaches an AI agent how to do one specific task. The agent sees only the skill's name and description until a task matches, then loads the full instructions.
- Agentic AI
- Agentic AI is the umbrella term for AI systems that take actions toward a goal with some independence, instead of only answering questions. It covers fixed AI workflows as well as AI agents that choose their own steps.
- AI agent
- An AI agent is an AI system that works toward a goal by choosing its own next steps: it picks a tool, uses it, looks at the result and repeats until the task is done. You set the goal and the limits, and the agent works out the path.
- AI eval (evaluation)
- An AI eval (evaluation) is a repeatable test that checks whether an AI prompt or workflow gives good answers, by running it on a fixed set of inputs and scoring the outputs against agreed criteria. It turns "seems fine" into a score you can compare.
- AI governance
- AI governance is the set of rules, owners and checks an organization uses to decide which AI uses are allowed, how their risks are handled, and who answers when something goes wrong. In the US, NIST publishes a voluntary framework for it, the AI Risk Management Framework.
- AI guardrails
- AI guardrails are checks and limits placed around an AI system that block or flag unsafe or wrong inputs and outputs before they cause harm. The strongest ones run as code outside the model, so a cleverly worded request can't talk its way past them.
- AI image generation
- AI image generation is the use of an AI model to create a new picture from a written description, or to edit an existing image by describing the change you want. You describe the subject, style and format in plain words, then refine the result in follow-up messages.
- AI memory (chat memory)
- AI memory is a chatbot feature that carries details from past conversations into new ones, such as your job, preferences or ongoing projects, so you don't have to repeat them. You can usually view, edit or delete what it has stored, or turn it off.
- AI observability
- AI observability is the practice of recording what an AI system does on every run, including each model call, tool call, cost and time taken, so you can see why it behaved the way it did. It is how teams debug and monitor AI once it runs on real work.
- AI orchestration
- AI orchestration is coordinating the steps around an AI model so a task runs reliably from start to finish: what starts it, which tools and data it gets, what happens when a step fails, and when a person must approve. The model is one step in that chain.
- AI ROI
- AI ROI (return on investment) is the value an AI project produces, in time saved, revenue, cost or quality, compared with what it costs to build and run. Measuring it properly requires a baseline recorded before the project starts.
- AI training opt-out
- AI training opt-out is a privacy setting that stops an AI provider from using your conversations to train its future models. In consumer chat apps it is usually a toggle in the account's data or privacy settings.
- AI vision (image understanding)
- AI vision is the ability of an AI model to look at an image you give it, such as a photo, screenshot, chart or scanned document, and describe it, read the text in it or answer questions about it.
- AI workflow
- An AI workflow is a fixed series of steps, set in advance by a person or by code, in which an AI model handles one or more of the steps. The steps run in the same order every time, which makes a workflow easy to test and check.
- API (application programming interface)
- An API (application programming interface) is a set of requests a software service agrees to accept from other programs, a bit like a menu. A program sends a request such as "create a contact" and the service sends back data or a confirmation.
- API key
- An API key is a long secret code that tells a software service which account is making a request. Anyone who has the key can act as you, within what that key is allowed to do, so it needs the same care as a password.
B
- Browser agent
- A browser agent is an AI agent that uses a web browser on your behalf: it reads pages, clicks, fills in forms and moves between tabs to complete a task. It can take over web chores you would otherwise click through by hand.
- Build vs buy (AI)
- Build vs buy is the decision whether to create an AI tool in-house or pay for an existing product or specialist vendor. Buying is usually faster and safer; building makes sense mainly when the workflow is a real competitive edge that no vendor covers.
C
- Centaurs and cyborgs
- Centaurs and cyborgs are two ways people work with AI. Centaurs split a task, handing some parts to AI and keeping others, while cyborgs work with AI continuously, passing small pieces back and forth.
- Chain-of-thought prompting
- Chain-of-thought prompting is asking an AI model to work through a problem step by step before giving its final answer. Writing out the steps improves accuracy on multi-step tasks like math and logic problems, and shows the working so you can check it.
- Chunking
- Chunking is splitting long documents into smaller pieces before they go into a RAG system, so a search can find and return just the relevant part. Where you cut the text decides whether each piece still makes sense on its own.
- Claude Code
- Claude Code is Anthropic's AI agent that works inside your files and terminal: it reads a project, edits files, runs commands and connects to other tools. Anthropic built it for software work, but it also handles other tasks made of files, APIs and repeatable steps.
- Claude Projects
- Claude Projects is a feature in Claude that groups chats into a workspace with its own instructions and uploaded files, so every conversation in that workspace starts already knowing your context.
- Coding agent
- A coding agent is an AI agent that works directly in a computer's files and command line: it reads, creates and edits files and runs commands based on plain-English instructions. Despite the name, it can do most file-based work, such as cleaning up spreadsheets or building a small app.
- Context engineering
- Context engineering is the practice of deciding what information an AI model sees each time it works: the instructions, documents, tool descriptions, memory and conversation history. It extends prompt engineering from the wording of one request to everything loaded around it.
- Context window
- A context window is the maximum amount of text an AI model can take into account at once, measured in tokens. It holds your messages, any files you add, its instructions and its own replies, and whatever doesn't fit is left out.
- Custom GPT
- A custom GPT is a version of ChatGPT set up for one job, with its own instructions, reference files and enabled tools, so anyone who uses it gets the same behavior without explaining the task each time.
D
- Deep research
- Deep research is a mode in AI assistants where the model runs many web searches on its own, reads the sources it finds and writes a longer report with citations. It takes longer than a normal answer and suits questions that need several sources compared.
- Digital twin (content)
- A digital twin, in AI content work, is a model of one person's writing voice, built from their past posts and notes, that drafts new content in their style. In engineering the term has an older meaning: a virtual copy of a physical object or system.
E
- Embedding
- An embedding is a list of numbers that represents the meaning of a piece of text, so a computer can measure how similar two texts are. Texts about the same idea get similar numbers even when they use different words.
- Error analysis
- Error analysis is the practice of reading a batch of real AI outputs, noting every way they go wrong, and grouping those failures by type before deciding what to fix or measure. It shows which problems actually happen, so your tests target those.
F
- Few-shot prompting
- Few-shot prompting is giving an AI model a few examples of the input and output you want, right inside the prompt, so it follows the pattern. The model picks up the format and style from the examples on the spot, with no retraining.
- Fine-tuning
- Fine-tuning is extra training that adjusts an existing AI model on your own examples so it learns a specific style, format or task. It changes the model itself, while a prompt or a document search only changes what the model sees for one request.
G
- Generative AI
- Generative AI is software that creates new content, such as text, images, audio, video or code, from a plain-language request. It learns patterns from large amounts of existing examples and produces fresh output that follows those patterns.
- Generative engine optimization (GEO)
- Generative engine optimization (GEO) is the practice of shaping content so AI answer engines, such as ChatGPT, Perplexity and Google AI Overviews, cite or mention it in their answers. It is the AI-search counterpart to SEO.
- Golden dataset
- A golden dataset is a small, fixed set of real example inputs, each paired with an answer you know is right, used to test an AI system. You run every new prompt or model against it and compare the results with those known-good answers.
H
- Hallucination
- A hallucination is a false statement an AI model presents as fact, such as a made-up statistic, quote or source. It happens because language models are trained to produce likely-sounding text, and likely-sounding text is sometimes wrong.
- Human in the loop
- Human in the loop is a setup where an AI system pauses at set points so a person can approve or correct its work before it takes effect. It is used for actions that are hard to undo, such as sending emails to customers or moving money.
J
- Jagged technological frontier
- The jagged technological frontier is the uneven edge of what AI does well: it improves results on some tasks and makes them worse on others, even when the tasks look equally hard. The term comes from a 2023 Harvard Business School study of consultants using AI.
K
- Knowledge cutoff
- A knowledge cutoff is the date after which an AI model has no information from its training. Anything that happened later, such as news, price changes or new product releases, is unknown to the model unless it searches the web or reads a document you give it.
L
- Large language model (LLM)
- A large language model (LLM) is an AI system trained on huge amounts of text to predict which words should come next. At scale, that one skill lets it answer questions, summarize, translate and write in a way that reads like a person wrote it.
- Latency (AI response time)
- Latency is the time an AI system takes to respond, from the moment a request is sent until the answer, or its first words, arrive. In agent workflows it adds up across every model call and tool call in a run.
- LLM-as-a-judge
- LLM-as-a-judge is a method of grading AI outputs by asking a second AI model to score them against written criteria, instead of having a person review every answer. It makes quality checks fast enough to run on every change, but the judge itself needs checking.
- Lost in the middle
- Lost in the middle is the tendency of AI language models to miss information placed in the middle of a long input while using information at the start or end more reliably. The name comes from a 2023 research paper by Liu and colleagues.
M
- MCP gateway
- An MCP gateway is a single entry point that sits between your AI agents and all of your MCP servers. It holds the passwords and keys, decides which tools each agent may use, and keeps one log of every call.
- MCP server
- An MCP server is a small program that connects an AI assistant to one tool or data source, such as Notion, Gmail or a database, using the Model Context Protocol. It tells the assistant what it can read or do there and carries out those actions when asked.
- Model Context Protocol (MCP)
- Model Context Protocol (MCP) is an open standard that lets AI apps connect to outside tools and data, such as a calendar, a CRM or a database, through one shared format. Apps that support MCP can then use those tools without a custom integration for each one.
- Model deprecation
- Model deprecation is when an AI provider announces that a model or API will be retired and sets a date after which requests to it stop working. Until then it still runs, but users are expected to move to a recommended replacement.
- Model pinning
- Model pinning is the practice of naming one exact, fixed model version in your code or settings, instead of a general name the provider can point at a newer model. It stops the model from changing under you before you have tested the new one.
- Multimodal AI
- Multimodal AI is AI that can work with more than one kind of input or output, such as text, images, audio and video, in the same conversation. You can show it a photo or screenshot and ask about it, or describe a picture and get one made.
N
- n8n
- n8n is a workflow automation tool where you build automations by linking steps, called nodes, on a visual canvas, starting from a trigger such as a schedule or an incoming webhook. Workflows can include AI steps and can run on n8n's cloud or on your own server.
- No-code automation
- No-code automation is building workflows that move data between apps by connecting visual blocks instead of writing code. Each workflow starts with a trigger, like a new email, and runs a set of steps, like creating a task, every time that trigger fires.
P
- Prompt
- A prompt is the instruction or question you give an AI model to tell it what you want. It can be one line or a full brief with background, examples and the format you need back, and the answer is shaped by everything in it.
- Prompt caching
- Prompt caching is an AI API feature that stores the processed start of a prompt you send repeatedly, so later requests beginning with exactly the same text are cheaper and faster. Only the repeated beginning is reused; the answer is still written fresh each time.
- Prompt engineering
- Prompt engineering is the practice of writing and testing the instructions you give an AI model so it produces reliable, useful output. For everyday users it means writing clear briefs; for teams running AI in production it means treating prompts like versioned, tested code.
- Prompt injection
- Prompt injection is an attack where text an AI reads, such as an email, web page or file, contains instructions that make the model do something its user did not intend. OWASP ranks it first in its Top 10 risks for LLM applications.
- Prompt library
- A prompt library is a shared, organized collection of prompts that work well, saved with clear names and version notes so you and your team can reuse them instead of rewriting them in every chat.
R
- Reasoning model
- A reasoning model is an AI language model trained to work through a problem step by step before it gives a final answer. The extra thinking makes it slower and more expensive per answer, and noticeably better at math, coding and other multi-step tasks.
- Regression testing for AI
- Regression testing for AI is re-running a fixed set of test cases after any change to a prompt or model, to check that what worked before still works. It catches changes that raise the average score while breaking one specific case.
- Retrieval-augmented generation (RAG)
- Retrieval-augmented generation (RAG) is a way of getting an AI model to answer from your own documents. A search step first finds the passages that match the question, then the model writes its answer using those passages.
S
- Semantic search
- Semantic search is a way of searching by meaning instead of exact words, so it finds passages that say what you asked even when they use different wording. It usually works by comparing embeddings, which are lists of numbers that capture what a piece of text means.
- Shadow AI
- Shadow AI is the use of AI tools at work without the company's approval or knowledge, such as staff pasting client documents into a personal chatbot account. It creates data risk, and it also shows where people already find AI useful.
- Speech-to-text (AI transcription)
- Speech-to-text is technology that turns spoken audio into written text. AI transcription models do this for recordings such as meetings and calls, so the words can then be searched, summarized or turned into tasks.
- Structured output (JSON mode)
- Structured output is a way of getting an AI model to return its answer in a fixed, machine-readable format, usually JSON with named fields, instead of free text. It lets another program read the answer reliably, for example to fill a spreadsheet row or pick the next step in a workflow.
- Subagent
- A subagent is a helper AI agent that a main agent hands a side task to: it works in its own separate context, with its own instructions and tool access, and sends back only a summary. This keeps the main conversation focused and lets tasks run in parallel.
- System prompt
- A system prompt is a set of standing instructions given to an AI model before the conversation starts, telling it how to behave in every reply. It sets the model's role, tone, rules and output format, and people chatting with the assistant usually never see it.
T
- Token
- A token is the small chunk of text an AI language model reads and writes, usually a short word or a piece of a longer one. In English one token averages about three-quarters of a word, and context windows and API prices are counted in tokens.
- Tool use (function calling)
- Tool use, also called function calling, is when an AI model asks your software to run a specific action, such as a search or a database lookup, and then uses the result in its answer. The model chooses the tool and the inputs; your code runs it.
- Trace
- A trace is the step-by-step record of a single run of an AI workflow or agent, showing every model call and tool call in order with its inputs, outputs, errors and timing. Reading one shows exactly where a run went wrong.
V
- Vector database
- A vector database is a database built to store embeddings, the number lists that represent the meaning of text, and to quickly find the ones closest to a new question. In many RAG systems it is the place where the search step happens.
W
- Webhook
- A webhook is a message one app sends to a web address you choose the moment something happens, such as a form being submitted or a payment going through. It lets another system react straight away instead of checking for changes on a timer.