AI Definitions

A way for software to learn from examples instead of being manually programmed for every rule.

An AI system that is very good at writing, explaining, summarizing, and transforming text.

AI that makes a new draft or artifact instead of only finding or sorting information.

The request you give AI.

Writing better AI instructions so you get useful, reviewable results.

The behind-the-scenes instructions that tell the AI how it should behave.

Showing the AI examples of what good looks like before asking it to do the task.

Asking AI to think through a problem step by step.

A creativity dial for AI responses.

AI answers in a clean format that another system can use.

Setting up everything the AI needs around the prompt, not just writing the prompt itself.

How much the AI can keep in view while answering.

A small piece of text the AI reads or writes.

When AI makes something up or gets a fact wrong.

AI looks up approved information before drafting an answer.

A common plug-in style connection layer that lets AI safely access approved tools and information.

The connector that makes a specific tool or data source available to AI.

The AI app that can use MCP-connected tools.

A specific action the AI is allowed to ask a connected system to perform.

Approved information the AI can read through an MCP connection.

A prebuilt instruction template the AI can use for a known task.

A bridge between AI and another system.

A way to turn meaning into numbers so computers can compare ideas.

A search system that finds similar meaning, not just matching keywords.

Search that understands what you mean, not just the words you typed.

Breaking big documents into AI-sized pieces.

A map of how important things relate to each other.

Teaching an existing model to behave more like you need for a specific pattern.

Making AI stick to approved facts instead of guessing.

Rules and controls that help keep AI use safe.

Sensitive information ending up somewhere it should not.

Trying to break the AI before real users do.

A trick used to get AI to ignore its rules.

When hidden or sneaky text tries to trick the AI into ignoring rules.

When the AI gets tricked by instructions hidden in something it was asked to read.

Knowing how sensitive information is before using or sharing it.

Information that points to a real person.

Information that needs extra care.

Give AI only the access it truly needs.

A person checks the AI before it matters.

AI that can do a multi-step workflow, not just answer once.

A multi-step AI workflow that can adjust as it goes.

AI can use a connected tool instead of only writing text.

AI fills out the right fields so software can run a specific action.

A software-to-software connection.

Managing all the moving parts of an AI workflow.

Turning repeated work into a guided or automated flow.

A required human yes before AI moves forward.

Things the AI remembers for later.

A trail of what happened, who did it, and what the AI touched.

Checking whether AI is good enough and safe enough for the job.

A test set for checking whether AI behaves the way you need.

The answer you consider correct for testing.

A check that confirms updates did not break old functionality.

A common test used to compare AI systems.

When AI flags something, how often is it right?

How much of the important stuff did AI catch?

One score that balances being right with not missing important things.

How long the AI takes to answer.

A speed limit for how much you can ask the system to do at once.

How AI usage turns into cost.

When AI gets less useful because the world or workflow changed.