AI glossary

AI that does the task instead of describing how.

An agent's steps drawn as boxes and arrows, with branches and loops.

The adjective on every AI product page. It means "it can act on its own."

Low-quality AI content made in bulk with no human judgment behind it.

The company behind Claude, Claude Code, and the MCP standard.

Anthropic's agent that works on your computer, not just in a chat.

OpenAI's coding agent. The ChatGPT-side answer to Claude Code.

Deciding what the model sees before you ask it anything.

A knowledge graph that also remembers decisions, rules, and when things were true.

How much text an AI model can work with at once.

A code editor with the AI agent built in.

OpenAI's always-on agents that keep working in the background toward goals you set.

AI that makes new things, not just sorts existing ones.

SpaceXSI's always-on agents that get their own computer and sign into your tools.

When the model states something false with total confidence.

The software around the model that makes it an agent.

The date the model stopped learning. Ask about anything after it and you get a guess.

A map of things and how they connect, so AI works from facts instead of loose text.

The kind of model behind ChatGPT, Claude, and Gemini.

Designing the loop that prompts the agent, so you don't have to.

The plug standard for connecting AI tools to your stuff.

The engine behind the chat. Not the app.

A model that takes images, audio, or video, not just text.

The company behind ChatGPT, the GPT models, Codex, and Dots.

Slang for "converted." Claude-pilled, agent-pilled, AI-pilled.

A folder in Claude with its own files, instructions, and chats.

Whatever you type in the box. All of it.

Hidden instructions in a web page or email that hijack your AI.

A model that thinks before it answers. Slower, pricier, better at hard problems.

Reusable instruction packs an agent loads when the task calls for them.

Standing instructions the model reads before you say anything.

Knowing good from plausible, now that making things is cheap.

The unit AI models count text in. Not words.

How a model asks the app to do something in the real world.

What the model learned from before you ever typed anything.

Telling an AI what you want and shipping the code without reading it.