Glossary

AI terms, in plain English.

49 words you'll hear in conversations about AI, explained without the hype. Written for business owners and managers, not engineers.

49 terms

A

Agent AI agent: An AI system that can take actions toward a goal, such as looking up records, sending messages or updating a system, rather than only answering questions. Agents need clear limits and checkpoints. See custom AI tools and agents.
AI use policy: A written set of rules for how employees may use AI tools at work: which tools are approved, what data can go into them, and when human review is required. See our guide.
API application programming interface: A defined way for one piece of software to talk to another. Most integrations and automations are built on APIs.
Artificial intelligence AI: A broad term for software that performs tasks normally associated with human intelligence, such as understanding language, recognizing patterns or making predictions.
Audit trail: A record of what a system saw, decided and did, and when. Important for reviewing automated actions and for compliance.
Automation: Software that performs a task without a person doing each step. Many valuable automations use simple rules; AI is useful when the input is messy, like emails or documents.

Bias: Systematic errors in an AI system's outputs that unfairly favor or disadvantage certain groups or outcomes, often inherited from training data.
Business associate agreement BAA: Under HIPAA, a contract required between a covered entity and a vendor that handles protected health information on its behalf.

Chatbot: Software that holds a text or voice conversation with a person. Modern chatbots are often powered by large language models.
Classification: Sorting inputs into categories, such as tagging incoming emails by topic or urgency. A common, reliable use of AI.
Context window: The amount of text a language model can consider at one time, including your instructions, any documents provided and the conversation so far.
Copilot: An AI assistant built into software you already use that helps with tasks like drafting, summarizing or analysis while you stay in control.

Data governance: The policies and responsibilities for how data is collected, stored, accessed, used and retired within an organization.
Data readiness: How prepared your data is for AI: whether it's accessible, reasonably accurate, consistently structured and allowed to be used for the purpose. See our guide.
Deterministic: Producing the same output every time for the same input. Traditional software is deterministic; language models usually are not.

Embedding: A list of numbers that represents the meaning of a piece of text, used to find related content by meaning rather than exact words.
Evaluation evals: Testing an AI system against a set of real examples with known good answers to measure how well it performs before and after changes.
Extraction: Pulling specific fields out of unstructured content, such as dates, amounts and names from invoices or forms.

Fine-tuning: Further training an existing AI model on your own examples to adjust its behavior. Often unnecessary; good instructions and retrieval solve many problems more cheaply.
Foundation model: A large AI model trained on broad data that can be adapted to many tasks. Most business AI tools are built on top of one.

Generative AI: AI that creates new content, such as text, images, audio or code, based on patterns learned from training data.
Grounding: Connecting an AI model's answers to specific, trusted sources, such as your documents or database, to make answers more accurate and verifiable.
Guardrails: Rules and checks placed around an AI system to prevent unwanted behavior, such as blocking certain topics, enforcing approval steps or honoring opt-outs.

Hallucination: When an AI model produces information that sounds plausible but is false or unsupported. The main reason human review matters.
Human in the loop: A design where a person reviews, approves or can override an AI system's output before it has real-world effect.

Inference: Running a trained AI model to produce an output. Usage-based AI pricing is usually for inference.
Integration: Connecting two or more software systems so data moves between them automatically. See data and systems integration.

Large language model LLM: An AI model trained on large amounts of text to understand and generate language. The technology behind most modern AI assistants.
Latency: How long a system takes to respond. Matters for customer-facing tools like chat and voice.

Machine learning ML: A branch of AI in which systems learn patterns from data rather than following explicitly programmed rules.
Model: The trained AI system that turns inputs into outputs. Different models vary in capability, speed, cost and data terms.
Multimodal: Able to work with more than one type of input or output, such as text, images and audio.

OCR optical character recognition: Technology that converts images of text, like scanned documents, into machine-readable text.
Open-weight model: An AI model whose trained parameters are published so organizations can run it on their own infrastructure, subject to its license.

PII personally identifiable information: Information that can identify a specific person, such as names, addresses, Social Security numbers or account numbers.
Prompt: The instructions and content given to an AI model to produce a response.
Prompt injection: An attack where text hidden in an input, such as an email or web page, tries to override an AI system's instructions. Listed in the OWASP Top 10 for LLM Applications.
Proof of concept POC: A small, quick build to test whether an idea works before investing in a full solution.

RAG retrieval-augmented generation: A technique where relevant documents are retrieved and given to a language model along with the question, so its answer draws on your information.
Robotic process automation RPA: Software that mimics clicks and keystrokes to automate tasks in applications, often used when no API is available.
ROI return on investment: The value gained compared to the cost. For AI, measure against a baseline you recorded before starting. See our guide.

Shadow AI: Employees using AI tools that haven't been approved by the business, often with company data. A common reason to write an AI use policy.
SSO single sign-on: Signing in to many applications with one company account. Makes it easier to control who can access AI tools.
Structured data: Information organized in a consistent format, such as rows and columns in a database. Easier for software to use than free-form text.
System of record: The authoritative source for a type of information, such as a CRM for customer data or an ERP for orders.

Token: A chunk of text, often part of a word, that language models process. AI usage is typically priced and limited by tokens.
Training data: The data used to train an AI model. Its quality and scope shape what the model does well and poorly.

Vector database: A database designed to store embeddings and quickly find items with similar meaning. Often used in RAG systems.

Workflow: A repeatable sequence of steps to get a piece of work done, such as onboarding a client or processing an invoice. Where most automation value lives.

Free 30-minute call

Tell us how your team works today. We'll tell you plainly where technology can save time, where it can't, and what a sensible first step would be.

Book a free call

Prefer the phone? Call (949) 691-0086