258 terms · 6 categories
Glossary
Every term comes with a plain-English definition, a wireframe, and a prompt you can copy straight into your AI.
Look-alike terms
23 sets of look-alike terms that are easy to mix up, side by side.
AI coding
What people say when working with AI coding tools, sorted A–Z.
A terms
- AI AgentAn AI that breaks down tasks, uses tools, checks the results, and decides what to do next on its own, instead of just replying with text one question at a time.
- AI AutocompleteThe editor suggests the next stretch of code in faint preview text as you type, and you accept it with Tab.
- AI EvaluationUse a fixed set of cases and scoring rules to see whether a prompt or model change got better or worse—not just two gut-feel tries.
- AI MemoryStore preferences and facts that should survive across chats and inject them next time, so you don't repeat yourself every session.
- AI Pair ProgrammingTreat the AI as a pairing partner: one of you drives writing and trying, the other (or the AI) questions, adds tests, and watches risk—instead of tossing the whole job to the model.
- API KeyA secret string that identifies and bills you when you call a large model or third-party service. It's basically your account password. If it leaks, other people can spend your money.
C terms
- CheckpointA snapshot of your code that an AI coding tool saves automatically before each round of changes. If something breaks, you can go back to before that round in one click.
- Claude CodeA coding agent from Anthropic that mainly runs in the terminal. It can understand the whole project, edit files, and run commands.
- Codebase IndexingThe AI editor scans and indexes your project ahead of time so chat and completions can find relevant code without rereading everything each turn.
- Computer UseLet the AI look at the screen, move the pointer, click, and type—operating a browser or desktop app the way a person would.
- Context EngineeringDeliberately choose and organize what goes into the model context—code, rules, retrieval hits, history—so the AI does the right work inside a limited window.
- Context WindowThe limit on everything a model can "see" in one conversation, measured in tokens. Anything over the limit gets cut off or compressed.
- CursorA code editor with built-in AI. It looks and works a lot like VS Code, and you can chat with the AI right in the editor and have it edit your project's code directly.
D terms
E terms
F terms
H terms
- HallucinationWhen AI generates content that looks reasonable but is wrong or doesn't exist, like made-up functions, API fields, or package names.
- Human-in-the-loopThe AI may propose or prepare actions, but critical steps wait for a human before they actually run—so it doesn't wreck production or delete data on its own.
I terms
L terms
M terms
- Model Context ProtocolAn open protocol that lets AI tools connect to external tools and data sources in a standard way. Install the right MCP server and the AI gets new abilities.
- Multi-file EditIn one task, let the AI edit several related files and show Diffs together, instead of pasting files back and forth in chat.
- MultimodalThe model can handle more than text. It can understand images, screenshots, design mockups, and other kinds of input, and some can also generate images or audio.
P terms
- Plan ModeHave the AI output a change plan, file list, and risks first, and only edit code after you approve.
- PromptWhat you send to the AI, including the request, background info, and requirements. The clearer and more specific it is, the closer the output gets to what you expect.
- Prompt InjectionAn attacker hides instructions inside user input or web content to make the model ignore its rules, leak the system prompt, or take dangerous actions.
R terms
- Rate LimitThe service limits how many requests you can send or how many tokens you can use in a period of time. Go over and it temporarily turns you away, usually with a 429 error code.
- RefactoringChanging how code is structured and written without changing what it does or how it looks, so it's clearer and easier to maintain and build on.
- Retrieval-Augmented GenerationFirst search a knowledge base for content related to the question, then put it in the prompt for the model to use. This cuts down on made-up answers and lets the model use up-to-date or private material.
- Rules FileLong-term agreements that live in your project and are given to the AI automatically in every conversation, like the tech stack, code style, and things not to do.
S terms
- Sandboxed ExecutionRun AI-generated code or commands in an isolated environment with limited network, filesystem, and secret access to reduce accidents and supply-chain risk.
- Screenshot to CodeSend the AI a design or a screenshot of a web page, and it picks out the layout and components and generates frontend code that looks similar.
- Spec-Driven DevelopmentFirst write the requirements, design, and task list into a document together with the AI. Once you approve it, build it item by item from the list.
- StreamingSend tokens to the UI as soon as they are generated so people can start reading before the full answer is done.
- Structured OutputRequire the model to return a agreed JSON/Schema so programs can parse it directly instead of scraping free-form text.
- System PromptUnderlying instructions given to the model before the conversation starts. They set its role, behavior, and rules, and usually take priority over normal messages.
T terms
- TemperatureA setting that controls how random the model's output is. Low values give steadier, more conservative answers. High values give more varied, creative ones, but they're also more likely to go off track.
- TokenThe smallest unit a large model reads and writes. How much the model can read and how much a call costs are both counted in tokens.
- Tool CallingWhile answering, the model asks to use external tools, like reading files, running commands, or searching the web, then keeps working based on what the tools return.
V terms
- v0An AI UI generator from Vercel. Describe what you want in words or upload a screenshot, and it generates React pages and component code you can preview.
- Vibe CodingA way of building software where you describe what you want in plain language, AI writes most of the code, and your job is mainly to ask, check the result, and give feedback.