Glossary · A to Z
AI glossary: plain words for the whole wheel.
Each term is defined as the books use it, with the chapter where it goes deeper.
Updated
73 terms
A
- A2A (Agent2Agent)An open protocol for agents from different vendors to find each other and pass work.
- AccountabilityA specific, named human is responsible for what an AI system does.
- ADLC (Agentic Development Life Cycle)A ten-phase method for taking a product to production with agents doing most of the building.
- AgentAn AI system that takes a goal and acts on it, using tools in a loop.
- Agent Accountability RegisterOne row per live agent: owner, permissions, traceability, escalation, kill switch tested.
- Agent Operating ModelA one-page contract for how one agent may behave.
- Agentic AIThe third stage: AI built from agents that plan and carry out work. The value frontier.
- Architecture (AI architecture)How the pieces of an AI system connect: to data, to tools and to the model.
- Autonomous vehicle (AV)A vehicle that drives itself using AI and sensors.
- Autonomy boundaryA set limit on what an AI system may do on its own.
B
C
D
- Data localisation (data residency)Legal rules that require certain data to be stored and processed inside a country's borders.
- Data readinessWhether the data an effort needs exists and is usable.
- Data sovereigntyData is subject to the laws of the place where it is collected and stored.
- Demo-grade vs production-gradeWorks once under perfect conditions, versus works reliably every day in real work.
- Deterministic guardrailA hard, rule-based limit an agent cannot cross, whatever its reasoning suggests.
- Digital twinA detailed virtual model of a real object or process.
- Disclosure (of AI)Being open about when a person is dealing with an AI system rather than a human.
- DRIVEThe five-phase method for turning an AI opportunity into value.
E
F
G
H
I
K
L
M
- Maturity model (AI Flywheel Maturity Model)Five levels, from Absent to Compounding, across three pillars: Stages, DRIVE and CORE.
- MCP (Model Context Protocol)An open standard for how AI systems connect to tools and data.
- Model (AI model)The trained system at the heart of an AI capability.
- Model card / Agent cardA standard document stating what an AI system is for, its limits and how to use it.
- Model censusA record of every model in use, each with an owner, card, licence, eval record and expiry date.
- MTBI (mean time between interventions)The average running time between moments a human has to step in.
- Multi-agent systemSoftware in which several specialised agents hand work to one another.
- MultimodalAn AI system that works with more than one kind of input or output.
O
P
R
- Reset taxThe hidden cost of treating each AI effort as a separate one-off.
- Rights ledgerA record of every input to a generative pipeline, with its source, licence and approval.
- RoboticsMachines that sense, decide and act in the physical world.
- ROI (return on investment)The value returned compared with the cost.
S
- Sim-to-real gapThe difference between how a skill scores in simulation and in reality.
- SimulationTeaching a robot inside a virtual world before it acts in reality.
- Sovereign AIKeeping data, models and decisions inside a nation's own legal jurisdiction.
- Standardisation (reusable building block)Turning a proven win into a documented, governed, reusable component.
- Success definitionA written statement of what an effort must achieve, set before building.
- Synthetic dataData made artificially, often from simulation.
T
V
- ValidationProving that an effort worked, against its success definition.
- Vanity metricA number that measures activity, not value.
- Vendor-neutralDesigned so that no single vendor's products are essential.
- VLA (vision-language-action model)A model that links what a machine sees and is told to the physical actions it takes.
W
Quick answers
Common AI terms, answered.
What is agentic AI?
Agentic AI is the third stage of AI: systems built from agents that plan and carry out work, using tools in a loop. It is harder and slower than generative AI and gated by governance, which is what makes it defensible (The AI Flywheel, Ch 6).
What is MCP?
MCP, the Model Context Protocol, is an open standard for how AI systems connect to tools and data. Use MCP for tools and A2A for peer agents; together they are often called the HTTP of agents (The AI Flywheel, Ch 6).
What is a kill switch for AI agents?
A kill switch is a tested, pre-approved way to stop an agent quickly. If it has never been tested, it is folklore, not a control. Every row in the Agent Accountability Register records whether it was tested (The AI Flywheel, Ch 17).