Glossary — A
Back to glossary index · Master INDEX
Accessibility (a11y)
The practice of building software that works for everyone, including people who use screen readers, keyboards only, voice control, or who have low vision, hearing loss, or motor differences. The “a11y” is a numeronym — “a,” then 11 letters, then “y.” Concretely it means: meaningful HTML, sufficient color contrast, keyboard-navigable interfaces, alt text on images, and labels on form inputs.
See also: Accessibility (a11y), WCAG
Agent (AI)
An LLM wired up with the ability to plan, take actions, and observe results in a loop, rather than just producing a one-shot response. Claude Code is an agent: it reads your message, decides which tools to call (read files, edit, run commands), looks at the results, and decides what to do next. The difference between a chat assistant and an agent is that the agent can act on the world, not just describe what to do.
See also: Agents, Tool use, Claude Code
Algorithm
A specific sequence of steps for solving a problem. “Sort this list of numbers from smallest to largest” is a problem; “compare each pair of adjacent numbers and swap if they’re out of order, repeating until no swaps are needed” is an algorithm (specifically, bubble sort). Every program is built out of algorithms — most of them so basic you don’t think of them as such.
See also: Algorithms — intro, Big O
Anthropic
The AI safety company that makes Claude. Founded in 2021, based in San Francisco. Notable for the Constitutional AI training approach, the focus on safety research, and for being the company whose CLI (Claude Code) this encyclopedia is being built with.
See also: Claude Code, Claude models
Agentic AI
AI that doesn’t just respond to a single prompt but plans, takes actions, and iterates over multiple steps to accomplish a goal. Claude Code is agentic — it reads your request, decides what files to look at, runs commands, observes results, and adjusts. Devin, ChatGPT’s agent mode, and Claude’s computer use are all agentic. The shift from “AI that talks” to “AI that does” is the defining trend of 2024-2026.
See also: Agent (AI), Agents, Cognition AI (Devin)
AGI (Artificial General Intelligence)
Hypothetical AI as broadly capable as a human across any intellectual task. Distinguished from current narrow AI which is good at specific things. Whether we’re approaching AGI is hotly debated — frontier reasoning models (o3, Claude 4) achieve PhD-level scores on many benchmarks; whether that’s AGI depends on your definition. The stakes if AGI arrives are enormous; safety considerations significant.
See also: AGI explainer, AI safety primer
Alignment (AI)
The challenge of getting an AI system to pursue the goals and values its designers intended — not what they literally specified, but what they actually meant. “Make users happy” is easy to specify and hard to interpret correctly. RLHF (Reinforcement Learning from Human Feedback) and Constitutional AI are alignment techniques. Misalignment can be subtle (AI optimises for the wrong proxy) or dramatic (AI pursues goals incompatible with human interests).
See also: AI safety primer, RLHF
API (Application Programming Interface)
A defined way for one piece of software to talk to another. When your frontend calls a “GET /users/123” endpoint on your backend, that’s an API call. When the Anthropic SDK sends a request to Claude, that’s an API call. Most modern APIs are HTTP-based and exchange JSON. “API” is also used loosely to mean any function or method exposed by a library — e.g. “the React API.”
See also: APIs overview, REST, GraphQL
ARIA (Accessible Rich Internet Applications)
A set of HTML attributes (like aria-label, aria-hidden, role) that tell assistive technologies (mainly screen readers) what an element is or does when the HTML itself isn’t enough. The first rule of ARIA is: don’t use ARIA. If a native HTML element does the job, use it. ARIA is a patch for cases where you must use a <div> to do something a <button> would normally do.
See also: Accessibility (a11y)
Array
An ordered list of values. In JavaScript: [1, 2, 3] or ["red", "blue", "green"]. You can read by position (arr[0] is the first element), add to the end (arr.push(...)), iterate with .map(), .filter(), .forEach(). One of the two most fundamental data structures in JavaScript — the other being object.
See also: Data structures, JavaScript
Async (asynchronous)
Code that doesn’t wait for slow operations to finish before moving on. In JavaScript, anything network-related (fetch, database queries, file reads in Node) is async. You handle async code with Promises and async/await. The opposite is synchronous — code that runs top-to-bottom, one line at a time, blocking on each.
const data = await fetch("/api/users").then(r => r.json());See also: Async & concurrency, Promise
Authentication (authn)
Proving who you are. Logging in with a password, scanning a fingerprint, clicking a magic link in an email — all authentication. Often abbreviated “authn” to distinguish it from authorization. The two are constantly confused even by experts.
See also: Authentication vs authorization
Authorization (authz)
Determining what you’re allowed to do once you’re logged in. A logged-in user might be allowed to see their own data but not someone else’s. Authorization is the layer that enforces those rules. Often abbreviated “authz.”
See also: Authentication vs authorization, Row-Level Security