AI terms,
served simply.

When computers are designed to learn, decide, create, or solve problems that normally require human intelligence.

A way for computers to learn patterns from data instead of following only hand-written rules.

A type of machine learning that uses many-layered neural networks to find complex patterns in text, images, or sound.

A system of connected mathematical units that work together to recognize patterns.

An organized collection of examples—such as text, images, or numbers—used to train or test AI.

The phase when an AI model studies examples and adjusts itself to make better predictions.

The moment a trained model uses what it learned to answer a new question or make a prediction.

Training with labeled examples so the model learns which answer belongs with each input.

Training with unlabeled data so the model discovers patterns or groups on its own.

Learning through trial and error, using rewards or penalties to guide better actions.

An AI model that predicts how an environment may change after an action, allowing it to simulate what could happen next.

AI that creates new content such as text, images, code, music, or video.

A mathematical system trained on data to perform tasks such as writing, translating, or classifying.

An application or service that lets people use an AI model to complete a task.

An AI system that understands everyday language and responds in a conversation.

A model trained on huge amounts of text so it can predict language and generate fluent responses.

The everyday way people speak and write, rather than computer code.

The message, question, or instructions you give an AI model to explain what you want.

The text-entry area in an AI application where you type a question, instruction, or request for the AI to respond to.

The answer or content an AI returns after receiving a prompt.

An AI response that sounds confident but contains incorrect, misleading, or invented information.

A small piece of text—a word or part of a word—that a language model reads and generates.

The maximum amount of information a model can consider at one time.

Extra focused training that adapts a general model for a particular task, subject, or style.

The first large training stage where a model learns broad patterns from a huge dataset.

Using knowledge learned for one task to improve performance on a related new task.

Giving an AI trusted, relevant information such as documents, database records, or search results so its answer stays connected to real sources.

A method where AI retrieves relevant documents before generating an answer grounded in those sources.

AI models whose trained weights are available for developers to download, inspect, adapt, and run, depending on the model’s license.

The set of technologies that work together to build and run a digital product, such as an AI model, database, app, and cloud service.

Anthropic’s family of AI assistants for conversation, writing, analysis, coding, and tool use.

Names used for different Claude model tiers, balancing capability, speed, and cost.

A conversational interface where you send messages to an AI assistant and receive responses.

Organized workspaces that group related conversations, files, and instructions around one goal.

A collaborative mode where Claude can help carry out organized, multi-step work like a teammate.

A coding environment where Claude can inspect, build, test, and debug software with tools.

A design-focused workflow for creating layouts, visuals, and web experiences with AI assistance.

Ways Claude can assist directly inside spreadsheets, presentations, and browser workflows.

Documents, code, applications, or designs created in a separate workspace beside the conversation.

A lightweight text format that uses simple symbols for headings, lists, links, and emphasis.

Saved writing preferences that guide an assistant’s tone, structure, and voice.

Information an AI assistant is allowed to retain to personalize future conversations and work.

Specific guidance that tells an AI how to behave for a particular project or workflow.

High-level preferences that shape how an AI assistant responds across tasks.

Reusable instructions or workflows that help an AI perform a repeatable kind of task.

Packages that add focused capabilities, tools, connections, or workflows to an AI system.

Bridges that let an AI system access approved information or actions in another service.

A capability that lets an AI interact with visible software by clicking, typing, and navigating.

A mobile workflow for sending and following tasks from a phone to a larger AI workspace.

A mode that gives an AI more reasoning time for difficult or multi-step problems.

A workflow where AI gathers information from multiple sources and creates a structured report.

Using live internet results to add current information beyond a model’s stored training knowledge.

AI jobs configured to run later or repeat on a timetable.

OpenAI’s conversational AI product for asking questions, creating content, analyzing information, and using tools.

Always-on AI agents in ChatGPT that can keep working on goals between conversations and ask you to review results or make decisions.

OpenAI's AI coding agent that helps build software by reading, writing, reviewing, and debugging code.

A family of language models pre-trained on large datasets to understand and generate language.

Generations of GPT models with different levels of reasoning, capability, speed, and tool use.

The AI research and product company that develops GPT models, ChatGPT, and related tools.

A developer interface for experimenting with model prompts, settings, and responses.

The text units used to measure model input, output, cost, and context length.

A setting that influences how predictable or varied a model’s next-word choices may be.

A training approach where human preferences help guide a model toward more useful and safer responses.

Improving prompts through testing so outputs become more accurate, consistent, and useful.

Customized experiences built around a general model using extra instructions, knowledge, interfaces, or tools.

An AI system that can plan steps, use tools, and take actions toward a goal.

A chip designed for many calculations at once, making AI training and inference faster.

A processor designed by Google to accelerate machine-learning calculations.

Using rented computing power and storage over the internet instead of owning every server.

A web browser that runs on a remote server and is accessed over the internet, rather than running on your own device.

A capability that lets an AI or automation operate a web browser by opening pages, clicking buttons, typing into fields, and scrolling.

A defined way for software systems to send requests and exchange information.

Secret codes that let an app prove it has permission to use an API or online service.

Named, reusable pieces of code that perform specific actions; an AI agent can request a function by supplying the inputs it needs.

An open standard that lets an AI app connect to outside tools, data, and services through one consistent setup.

An open standard that lets independent AI agents discover one another, exchange messages, and work together across different systems.

A digital identity and set of permissions assigned to software, automation, or an AI agent instead of a person.

A searchable catalog that records which AI agents exist, who owns them, what they can do, and where they are in their life cycle.

An isolated workspace where an AI agent can use approved files, tools, and network connections without open access to the host system.

Software placed around a model to give it a friendlier interface, instructions, tools, or controls.

Systems that hold information an AI agent can retrieve, such as current records, past data, or reference documents.

A database that stores numerical representations so AI can quickly find similar content.

A coordinated set of tools and services that work together to deliver an AI solution.

Designing how to phrase a request, including clear instructions, examples, and an output format, to guide an AI toward the result you need.

A technique for getting more complex and nuanced results from large language models like Gemini by using the output of one prompt as input to the next. It is like a conversation where each response builds upon the previous one, helping produce a more sophisticated and refined outcome.

Selecting and organizing what goes into an AI model’s context window, such as instructions, documents, conversation history, and tool results, so it has the right information for the task.

Building the tools, guardrails, and saved progress around an AI model so an agent can act, follow boundaries, and resume work reliably.

Designing how one AI agent repeatedly chooses an action, uses a tool, checks the result, and continues until it finishes, reaches a limit, or needs help.

Designing how agents or workflow steps connect, share information, and hand off tasks to form a coordinated system.

Coordinating a multi-step process so people, software, or AI agents run in the right order, share results, and recover when a step fails.

An AI model built to choose from set options, answer yes or no, or assign a score instead of writing an open-ended response.

An engineer who works alongside a customer to build and deploy software for the customer's real workflows, often connecting AI to existing systems.

Applications that use two or more AI agents with different roles to work on parts of a task and bring their results together.

Saving an AI agent's progress so an interrupted multi-step run can resume safely without starting over or repeating completed work.

Tracking an AI agent's steps, tool use, errors, speed, quality signals, and cost so people can understand and improve its behavior.

Repeatable tests that check whether an AI agent completes tasks correctly and follows its rules before or while people use it.

When an organization creates AI agents faster than it can track their owners, purposes, permissions, costs, and results.

An interactive screen that an AI assembles for the current task instead of giving only a fixed page or plain-text answer.

Software whose source code is available under a license that lets people use, study, change, and share it.

A project folder on GitHub that stores a project’s code, files, and history in one shared place.

A saved checkpoint that records a specific set of changes to a project.

A separate version of a project where you can work on changes without disturbing the main version.

A request to review and merge your changes into the shared main project.

The front-page guide that explains what a project is, how to use it, and how to get started.

A tracked task, question, bug, or improvement that the project team needs to discuss or fix.

Automated jobs that run steps such as tests, builds, or deployments when a project changes.

An AI coding assistant that suggests code and explains options while you work.

Using an AI coding agent that can inspect files, edit code, run tools and tests, and correct its work across multiple steps.

An application you use through a web browser to do tasks, such as editing documents, booking appointments, or managing projects.

A private online space where clients sign in to view project updates, share files, send messages, or manage services with a business.

Techniques that help computers read, analyze, translate, and generate human language.

Making an AI system’s reasoning or decision factors understandable enough to review and question.

Work intended to make AI behavior better match human goals, values, instructions, and safety needs.

Technical limits that control what an AI agent can access, change, send, or spend while it works.

An operator-controlled way to quickly stop an AI agent's actions or access when it malfunctions or creates risk.

Checking whether a particular AI agent may take a particular action on a particular resource under the current rules.

Low-quality text, images, or video made with AI and published mainly for volume or attention.

A program that communicates through text or voice to answer questions or perform tasks.

An AI assistant controlled mainly through spoken questions and commands.

When an AI uses a web browser to navigate pages and take actions for a person instead of only returning search results.

Shopping or business buying in which an AI agent helps find, compare, or purchase goods and services on a person's or company's behalf.

An organized collection of trusted information that an AI system can search or reference.

Splitting text into smaller tokens so a language model can process it.

Numerical representations that place words with related meanings near one another.

When a model handles a new task without labeled examples of that exact task.

Keeping a person involved to guide, approve, or correct an AI system’s work.

An AI agent with a standing goal that can track progress and resume approved work after a conversation ends when a schedule, event, or new information calls for action. ChatGPT dots are one example of this type of agent.

Treating higher AI-token usage as a sign of productivity even when it does not improve the result.