AI Glossary.

When an AI generates information that sounds confident and credible but is factually wrong or completely made up. The model doesn’t “know” it’s wrong — it’s filling gaps with probability.

Why it matters: Hallucinations are invisible without verification. A well-governed AI flags uncertainty instead of guessing.

The input you give an AI to generate a response. Can be a question, instruction, context, or combination of all three. Quality of the prompt directly determines quality of the output.

Why it matters: Most AI failures start with a bad prompt, not a bad model.

The unit AI models use to process text. Roughly 1 token = ¾ of a word. Models have a maximum token limit per conversation — called a context window.

Why it matters: Long conversations or documents can exceed the context window, causing the AI to “forget” earlier content.

The maximum amount of text (measured in tokens) an AI can process in a single session. Everything outside the window is invisible to the model.

Why it matters: If your conversation exceeds the context window, the AI loses earlier instructions and starts drifting.

A setting that controls how creative or random an AI’s responses are. Low temperature = more predictable, focused output. High temperature = more varied, creative output.

Why it matters: For governed, reliable output — lower temperature is better. For brainstorming — higher works.

Instructions given to an AI before the conversation starts. Sets the rules, tone, role, and behavior for the entire session. This is what AI Blueprint™ generates.

Why it matters: A strong system prompt is the difference between consistent, governed output and generic, drifting responses.

Training a base AI model further on a specific dataset to make it better at a particular task or domain. Different from prompting — it changes the model itself, not just the instructions.

Why it matters: Fine-tuning is expensive and technical. Most operators don’t need it — a well-structured system prompt gets most of the way there.

Zero-shot means asking the AI to do something without giving it any examples. Few-shot means giving it 2-5 examples before the task to guide the format or style of the output.

Why it matters: Few-shot prompting dramatically improves output quality for complex or format-specific tasks.

A prompting technique that instructs the AI to reason through a problem step by step before giving a final answer. Reduces errors on complex reasoning tasks.

Why it matters: “Think step by step” is one of the most effective prompt additions for accuracy.

The process of running a trained AI model to generate a response. When you send a prompt, the model performs inference to produce output.

Why it matters: Inference speed and cost determine how practical a model is for real-time, high-volume use.

An AI system that can take actions autonomously — browsing the web, sending emails, executing code, or making decisions — without human input for each step.

Why it matters: Agents are powerful and risky. Without governance, they can take irreversible actions based on bad assumptions.

A sequence of AI-driven tasks where each step triggers the next — often without human review between steps. The AI plans, executes, and adjusts on its own.

Why it matters: Errors compound in agentic workflows. One bad assumption early can cascade into a significant problem downstream.

The ability of an AI to call external tools — search engines, APIs, calculators, databases — to complete a task. Also called “function calling.”

Why it matters: Tool use extends what AI can do but introduces new failure points — especially when tools return unexpected data.

The coordination of multiple AI agents or tools working together toward a shared goal. An orchestrator directs which agent does what and when.

Why it matters: Complex orchestration requires clear governance — who has authority, what can be done autonomously, and where humans must approve.

A system design where a human must review and approve AI output before it’s acted on or published. The opposite of full automation.

Why it matters: The right level of human oversight depends on stakes. Low-risk tasks can run autonomously. High-stakes decisions need human sign-off.

Autonomous AI acts without human review. Supervised AI requires human approval at defined checkpoints. Most real-world deployments sit somewhere in between.

Why it matters: The right balance depends on the task, the stakes, and how well the AI has been governed.

When an AI declines to answer or complete a task — either because it lacks sufficient information, the request violates its rules, or the output would be irresponsible.

Why it matters: In a governed system, refusal is a feature — not a failure. It means the system is working.

The human tendency to trust AI output because it sounds authoritative and professional — regardless of whether it’s accurate. Models are trained to sound confident even when wrong.

Why it matters: Confidence bias is one of the leading causes of AI-related mistakes. Governance systems are designed to counteract it.

A structured system for deciding which tasks to hand off to AI and which to keep human. Defines the boundaries of AI authority in a workflow.

Why it matters: Without a delegation framework, teams either under-use AI (doing everything manually) or over-delegate (creating unreviewed risk).

A documented, repeatable process for completing a specific task. When combined with AI, SOPs become the instruction layer that governs how AI executes recurring work.

Why it matters: AI Blueprint™ Business maps your SOPs into AI-ready instructions so every team member gets consistent output.

The practice of crafting prompts to get better output from AI. Focuses on the input — word choice, structure, examples, and constraints.

Why it matters: Prompt engineering is tactical. Prompt governance is strategic. iWasGonna™ focuses on governance — the system behind the prompts.

Using AI to execute a sequence of tasks automatically — triggered by an event, condition, or schedule — without requiring human input at each step.

Why it matters: Automation without governance creates invisible failure points. Every automated workflow needs defined checkpoints and refusal rules.

A technique where an AI retrieves relevant documents or data from an external source before generating a response — grounding its output in real, current information rather than training data alone.

Why it matters: RAG dramatically reduces hallucination for knowledge-intensive tasks by anchoring the AI to verified sources.

Mathematical representations of text (or other data) as vectors — lists of numbers that capture semantic meaning. Similar concepts have embeddings that are close together in vector space.

Why it matters: Embeddings are the foundation of semantic search and RAG systems. They let AI find relevant content by meaning, not just keyword match.

A database optimized for storing and searching embeddings. Used in RAG systems to quickly find the most semantically relevant documents for a given query.

Why it matters: Vector databases are the retrieval engine behind most production AI knowledge systems.

Search that finds results based on meaning and intent rather than exact keyword matches. “What causes AI to fail?” finds results about hallucination even if that word isn’t in the query.

Why it matters: Semantic search makes knowledge retrieval far more accurate and useful than traditional keyword search.

Breaking large documents into smaller pieces (chunks) before embedding them. The size and method of chunking affects how accurately an AI can retrieve relevant information.

Why it matters: Poor chunking leads to retrieval failures — the AI pulls the wrong section of a document and generates a misleading answer.

A secondary step after initial retrieval that re-scores and reorders results by relevance before passing them to the AI. Improves precision when the initial retrieval returns too many or noisy results.

Why it matters: Re-ranking is the difference between a RAG system that returns the right answer and one that returns adjacent-but-wrong information.

Retrieving the K most relevant documents or chunks before passing them to the AI for generation. K is a tunable parameter — higher K = more context, more noise. Lower K = less context, more precision.

Why it matters: Tuning K is a core tradeoff in RAG system design. Too low and the AI misses relevant context. Too high and it gets confused by irrelevant material.