The language of your next AI conversation — with your team, your vendors, and your competitors — including the terms that matter most in regulated settings like healthcare.
Agent / agentic AI — AI that carries out multi-step tasks — booking, researching, operating software — not just answering. The frontier of what can be automated.
Automation vs augmentation — Whether AI replaces a task outright or assists a person doing it — a strategic choice with very different risk and change-management profiles.
Compliance (e.g. PHI) — Meeting the rules that govern sensitive data and decisions — in healthcare, protected health information (PHI) and its handling sit at the centre of any AI use.
Data privacy / PII — Protecting personal and confidential information. Where data goes when staff paste it into AI tools is a legal and reputational question — acute for patient data in healthcare.
Deployment / wrapper — Most AI products are an interface and safety layer around a frontier model. Buy the workflow and support — but ask what's underneath, and on whose terms.
Fine-tuning — Further training a model on your data or for a specific behaviour — one way products and deployments are specialised.
Generative AI — AI that produces new content on demand — copy, images, code, analysis — the capability behind most current business use cases.
Governance — The policies, roles, and controls for adopting AI responsibly — covering risk, data, and accountability. What turns pilots into safe scale.
Guardrails — The product-level controls on what an AI will and won't do. A key thing to scrutinise before you deploy.
Hallucination — Confident, plausible, wrong output. The core reliability risk to design around — with grounding, review, and human sign-off.
Human-in-the-loop — Any workflow where AI proposes and a person decides. The accountability rule for consequential or regulated decisions.
LLM (large language model) — The engine inside AI assistants: trained on vast text to predict the next word. Most products you'll evaluate are built on one.
Open-weight model — A model you can download and run on your own infrastructure — relevant when privacy, cost, or control rule out sending data to a vendor.
RAG (retrieval-augmented generation) — Grounding AI answers in your own documents and data rather than the model's memory. The standard way to make AI reliable on your business.