Artificial Intelligence
Think of AI as a computer helper that can spot patterns, write drafts, answer questions, and make suggestions.
You still decide whether the answer is useful.AI jargons
AI words explained like you are hearing them for the first time.
If a meeting, article, or tool says a word that sounds bigger than it needs to be, start here. You will get the plain meaning, a tiny example, and the simple thing you should remember.
Start here
Basic AI words you will hear everywhere
These are the words you need before you read tool docs, compare assistants, or ask AI to help with work.
The engine behind the answer
A model is the part that learned from many examples and tries to predict a helpful answer for you.
Different models can be better at different jobs.The instruction you give AI
A prompt is what you ask for. A better prompt says what you want, what matters, and what should not change.
Do not just say “make it better.” Tell it what better means.The answer you get back
Output is the reply, code, summary, image idea, test plan, or explanation the AI gives you.
Treat output like a draft until you check it.A tiny piece of text
AI reads and writes in small pieces called tokens. More text usually means more tokens.
Too many tokens can cost more and make the assistant wander.The information AI sees
Context is the helpful background you give AI: files, goals, rules, examples, and constraints.
Good context is like giving a map before asking for directions.The AI backpack
The context window is how much information AI can carry in one task.
If the backpack is stuffed with junk, the useful things are harder to find.When AI makes the answer
Inference is the moment the model uses your prompt and context to produce a response.
You can think of it as “AI answering time.”When AI helps with software
Coding assistant words without the fog
You will hear these terms when you use Claude, Codex, Copilot, Cursor, Windsurf, Kiro, or an internal AI tool.
A helper for software work
A coding assistant can explain code, write a patch, suggest tests, or help you understand a bug.
It is a helper, not the final reviewer.An assistant that can take steps
An agent can plan, inspect files, run tools, and make changes across a task.
Give agents clear boundaries before they start moving around.House rules for AI
Repo instructions tell the assistant how your project works, which commands to run, and what rules to follow.
Files like AGENTS.md or CLAUDE.md help keep behavior consistent.The before-and-after view
A diff shows exactly what changed in the code.
Always review the diff before you trust an AI-generated change.The first structure
A scaffold is the starter shape of a feature, page, service, or test setup.
It is like a frame before the walls and wiring are finished.When old behavior breaks
A regression happens when a new change accidentally breaks something that used to work.
Brownfield work needs regression checks before release.AIDLC words
Workflow terms that help you use AI safely
These words turn AI from a guessing machine into a reviewable team workflow.
Why you are doing the work
Intent tells the assistant and your team the real reason for the change.
Example: “Reduce failed checkout retries by making the payment error clearer.”What not to touch
A non-goal protects your work from growing too wide.
Example: “Do not change payment provider behavior.”A structured request
A prompt contract says what you want, what context matters, what is allowed, and what evidence you need.
It keeps the assistant from guessing the rules.A checkpoint before trust
A review gate is a place where a human checks AI output before it moves forward.
Do not skip it just because the answer sounds confident.Proof the work is safe
Evidence can be tests, screenshots, review notes, logs, or acceptance decisions.
If you cannot show proof, the change is not ready.Your undo plan
Rollback is how you safely undo a change if it causes trouble.
Know the undo path before release, not after panic starts.Watch-outs
AI words that should make you slow down
These terms are not scary. They are reminders to check the work before you accept it.
A confident wrong answer
Hallucination is when AI says something that sounds real but is wrong, missing, or made up.
Ask for evidence and verify important claims.When AI does too much
This happens when you ask for one small change and the assistant rewrites more than needed.
Use scope boundaries and non-goals.Wasted AI usage
Token burn happens when vague prompts, huge context, and repeated retries spend usage without moving the work forward.
Start with intent and context before asking for output.