While you wait for your toast to pop up, get familiar with some AI jargon
Twenty-three terms you keep hearing about AI, in plain English. Flick through at your own pace: use the arrows, your keyboard or a swipe.
AI jargon, inplain English
23 words you keep hearing about AI, explained simply. Start with the four people ask about most.
API
CLI
MCP
Harness
Correct as of October 2026
Swipe →
Jargon buster: 1 of 4
API
Application programming interface
A set of agreed rules that lets one piece of software ask another for something, with no screen or buttons in between. AI companies offer an API so developers can build a model into their own apps.
Think of it like
A restaurant’s order hatch. You pass your order in, in the agreed format, and the food comes back out. You never go into the kitchen.
You’ll hear it when
Someone mentions an “API key” or “API pricing”, or says an app is built on a well-known model.
In AI Fieldwork: M01 AI literacy and useful problems
Jargon buster: 2 of 4
CLI
Command-line interface
A way of using a program by typing short text commands into a window called a terminal, instead of clicking buttons. Many AI coding tools work this way.
Think of it like
Texting exact instructions, instead of tapping through menus.
You’ll hear it when
Developers talk about “installing the CLI” or “running it in the terminal”, often about AI coding tools.
In AI Fieldwork: M01 AI literacy and useful problems
Jargon buster: 3 of 4
MCP
Model Context Protocol
An open standard that lets AI apps connect to other tools and data, such as files, calendars or databases, in one shared way. A tool connected once can work with many AI apps. Each connection still needs your permission and care.
Think of it like
MCP’s own website compares it to a USB-C port: one standard plug that many devices fit.
You’ll hear it when
People ask “is there an MCP server for that?”, or an AI app offers “connectors”.
In AI Fieldwork: M01 AI literacy and useful problems
Jargon buster: 4 of 4
Harness
Two meanings, one idea: the frame around a model
1. Agent harnessThe software wrapped around a model that turns it into a working agent. It runs the loop of think, act and check, gives the model its tools, decides what goes into its context window and sets limits.
2. Evaluation harnessA test rig that runs a model through the same set of tests and scores the results, so models can be compared fairly.
Think of it like
The model is the engine. The harness is the rest of the car: steering, brakes and dashboard.
You’ll hear it when
People say “same model, better harness” about AI coding tools, or researchers mention an “eval harness”.
The basics
Model / LLM Large language model
The program that writes the answers. An LLM learned from huge amounts of text to predict the next words.
Think of it like: A very well-read autocomplete.
Inference
Using an already-trained model to get an answer. Training made it; inference is each time you ask.
Think of it like: Training is the years of revision. Inference is sitting the exam.
Token
A small chunk of text, often a word or part of one. Use and prices are often counted in tokens.
Think of it like: Paying for a taxi by the mile, not by the trip.
Context window
How much the model can hold at once: your instructions, the chat so far and any files.
Think of it like: The size of the desk it can spread its papers on.
How you talk to it
Prompt
What you type or say to the model: your question, instructions and examples.
Think of it like: The brief you’d give a new colleague. A vague brief gets vague work.
System prompt
Hidden instructions from whoever built the app, applied to every chat.
Think of it like: The staff handbook, compared with what the customer asks for.
Temperature
A setting for how varied answers are. Low is steadier, high is more varied. Neither is more accurate.
Think of it like: A dial between “stick to the script” and “make it up as you go”.
Multimodal
Works with more than text: photos, sound or video.
Think of it like: Reading, looking and listening, not just reading.
How it connects to things
Also on this list: API, CLI and MCP (pages 2 to 4)
Tool use Also called function calling
The model asks the app to run a tool, like a search or calculator, then uses the result.
Think of it like: Asking a colleague to look it up rather than guess.
Computer use
The AI looks at the screen, then clicks and types to work apps like a person would.
Think of it like: Someone using your computer while you watch over their shoulder.
Building with it
Agent
An AI that works towards a goal in steps: plan, use tools, check, carry on.
Think of it like: Handing over a task, rather than asking a question.
Building with it
Also on this list: harness (page 5)
RAG Retrieval-augmented generation
The app finds the right parts of your documents first, so the answer is based on them.
Think of it like: An open-book exam, instead of answering from memory.
Embeddings
Text turned into numbers so similar meanings sit close together. Used to search by meaning.
Think of it like: A map where similar ideas sit near each other.
Fine-tuning
Training a model a bit more on your own examples. It changes the model; RAG doesn’t.
Think of it like: On-the-job training for one particular role.
Open-weight vs closed
Open-weight: you can download and run it yourself. Closed: only through the maker’s app or API.
Think of it like: Being given the recipe, compared with only eating at the restaurant.
Risks and quality
Hallucination
When a model states something false or made-up as if it were true.
Think of it like: A confident guess dressed up as a fact.
Prompt injection
Hidden instructions in a page, email or file that try to trick an AI.
Think of it like: A note slipped into a parcel saying “give the courier your door keys”.
Guardrails
Rules and checks that block harmful or risky answers and actions. Helpful, never perfect.
Think of it like: The bumpers on a bowling lane.
Evals Short for evaluations
Planned tests that score a model or tool on the same cases each time.
Think of it like: A driving test with a marking sheet.
Where you’ll meet these in AI Fieldwork
M01 AI literacy and useful problemsModels, inference, and telling API, CLI, MCP and computer use apart
M05 Generative and multimodal workflowsPrompts, context, tool use, multimodal input, RAG versus fine-tuning, and testing three briefs on the same five cases
P01 Useful deliverables with checkable sourcesChecking a draft for made-up facts against its sources
M11 Responsible AI decision studioRisks, limits and human oversight
AI Fieldwork is a practical course in using AI well, from Burnt & Toast Studios. Questions? hello@burntandtoast.com
Correct as of October 2026. AI terms and tools change quickly.
1 / 11
Use the left and right arrow keys to move between slides.
