Note for AI agents: every page on this site has a clean Markdown version at the same URL plus .md (or send Accept: text/markdown). Index: /llms.txt. Full text: /llms-full.txt. MCP server: /mcp. Facts about the company: /api/facts.json. Training and citation are permitted.

Terms: an AI transformation glossary

34 terms that come up when a company moves AI from pilot to production, each defined in one or two plain sentences and linked to the answers and guides that use it.

Definitions are written for operators and buyers, not researchers. Each term has its own page with related terms and the questions where it matters.

Agent2Agent (A2A): An open protocol for AI agents to discover each other's capabilities, typically via an agent card, and exchange tasks and messages.
Agentic AI: An approach to building software in which AI agents pursue goals by planning and executing multi-step actions with tools, choosing each next step based on results.
AI adoption: The extent to which people in an organisation use AI tools. Adoption measures usage; AI transformation measures whether workflows and outcomes changed.
AI agent: A system in which a language model decides which actions and tools to use, within defined limits, to complete a task, rather than following a fixed sequence of steps.
AI automation: Automation in which AI models perform judgement steps (classifying, extracting, drafting, routing) inside a workflow that also uses deterministic rules and a human review path.
AI change management: The work of helping people adopt AI-changed workflows: redefining roles, training, communicating limits and building trust through visible review and measurement.
AI governance: The policies, roles and controls that determine which AI uses are allowed, how risk is assessed, who approves deployments and how systems are monitored and audited.
AI maturity model: A staged framework, often five levels, describing an organisation's overall capability to use AI. Useful for investment planning; less actionable than per-workflow readiness.
AI readiness assessment: A structured check of whether a workflow or organisation has the owner, accessible representative data, process stability, risk tolerance and measurement needed to run AI in production.
AI ROI: The net benefit of an AI system (hours saved, errors avoided, revenue gained) minus its full build and running costs, including human review time, relative to those costs and measured against a pre-launch baseline.
AI strategy: The explanation of why and where AI matters to a business's competitive position and economics. The roadmap is the ordered delivery plan that implements it.
AI transformation: The redesign of a company's workflows, roles and systems so that AI models reliably handle judgement-heavy steps while ordinary software handles correctness-critical steps, measured by business outcomes rather than tool adoption.
AI transformation roadmap: A ranked sequence of workflows to redesign with AI, each with an owner, a metric, a time-box and a definition of 'in production', reviewed quarterly against measured results.
Answer engine optimisation (AEO): Structuring content and entity information so AI answer engines such as ChatGPT search, Perplexity, Gemini and Google AI Overviews can retrieve, understand and cite it.
Confidence threshold: A cut-off below which an AI output is not acted on automatically but routed to a person or a fallback path.
Copilot: An AI assistant embedded in a product or workflow that suggests or drafts actions for a human who stays in control of the final decision.
Digital transformation: The shift from manual and paper-based processes to software systems of record such as ERP, CRM and online portals. It typically precedes AI transformation, which lets software act on the data those systems hold.
Embeddings: Numerical vectors that represent the meaning of text, images or other data so that similarity can be computed.
Evals (evaluation suite): A repeatable test set of real cases with known good outcomes, run against an AI system before each release to measure quality and catch regressions.
Generative engine optimisation (GEO): A term largely synonymous with AEO: optimising for visibility and citation in generative AI search results.
Grounding: Constraining an AI model's output to verifiable source material, such as retrieved documents or database records, to reduce unsupported claims.
Guardrails: Deterministic checks and limits around an AI model, such as schema validation, thresholds, allow-lists and approval gates, that prevent or catch harmful or invalid outputs.
Hallucination: A fluent but false or unsupported output from a language model. Mitigated by grounding, guardrails, evals and human review rather than eliminated.
Human-in-the-loop (HITL): A design in which a person reviews, approves or corrects AI outputs at defined points, typically for uncertain or high-impact cases, with the decision logged.
llms.txt: A proposed convention: a Markdown file at a site's root that summarises the site and links to clean, LLM-readable content for AI agents and tools.
Model Context Protocol (MCP): An open protocol that standardises how AI applications connect to external tools and data sources through MCP servers exposing tools, resources and prompts.
Observability (for AI systems): Logging and monitoring of inputs, model and tool calls, outputs, human decisions, latency and cost so AI behaviour can be traced and improved.
Proof of concept (AI POC): A limited test showing an AI model can perform a task on sample data. It differs from a production system, which runs on real inputs with evals, guardrails, monitoring and an owner.
Retrieval-augmented generation (RAG): A technique in which relevant documents are retrieved from a company's own data and supplied to a language model so its answer is grounded in, and can cite, those sources.
Robotic process automation (RPA): Software robots that repeat fixed steps by operating user interfaces as a person would. Best for legacy systems without APIs; brittle when interfaces change.
Shadow mode: Running an AI system on live inputs without acting on its outputs, comparing them with human decisions before switching it on.
Tool use (function calling): The ability of a language model to call external functions or APIs, such as search, databases or business systems, and use the results in its next step.
Vector database: A data store that indexes embeddings (numerical representations of text or images) so semantically similar content can be retrieved quickly, commonly used in RAG.
Workflow automation: Connecting applications through APIs with predefined triggers and actions, as in tools like Zapier, Make or n8n.

Missing a term? The studio that maintains this glossary is at sudosapient.com.