AI news, product pages, and app settings use plenty of shorthand. You’ve met most of these words in earlier lessons, and a few more come up often in headlines. Each term below links to Real Python’s AI Coding Glossary where it has an entry, so you can hover over it for a quick reminder or open it to read more.
Words for the Parts
These terms name the parts of an AI system and the kinds of work it does:
- An LLM, short for large language model, is the general name for models like GPT, Claude, Gemini, and Grok.
- Training is how a model learned, by reading huge amounts of text and predicting what comes next. Training data is that text, and pretraining is the first, biggest round of it.
- The knowledge cutoff is the date a model’s training text ends. For anything after it, the app has to search.
- A harness is the app around the model. News stories often say “the app” or “the product” instead.
- An agent is a model and a harness working together on a task over many steps. Agentic describes work done that way, as in “agentic features.” Agentic coding is the same idea applied to writing code.
- Tool use, also called function calling, is when the model asks the app to take one action, like searching the web or opening a file.
- A reasoning model writes out more of its thinking before it answers. That’s what a “thinking” mode turns on, and it’s why those modes take longer and use more.
- A prompt is the request you send to the model. Prompt engineering means writing prompts carefully. For day-to-day work, it mostly comes down to what you practiced in Asking Well.
- A system prompt is the set of standing instructions the app gives the model before your first message.
- Vibe coding means building an app by describing it in plain words and asking for changes, often without reading the code. App builders are made for it.
