IAG Blog

Glossary

Working definitions for concepts used across the dev blog system. These terms support navigation and review; Scott can still revise the final editorial language.

Architecture

control plane

The policy, access, monitoring, evaluation, and cost-management layer that makes AI systems governable across teams and workflows.

Also referenced as: AI control plane

model portfolio

A managed set of models and providers selected for different cost, risk, capability, privacy, and workflow requirements.

Also referenced as: model choice, multi-model strategy

Data

data boundary

A defined rule for which company, customer, employee, or regulated data may be used with a particular AI system and under which controls.

Also referenced as: data boundaries, data-use boundary

Governance

agent governance

The controls used to define, test, monitor, and constrain AI assistants or agents before they are allowed to act inside real workflows.

Also referenced as: assistant governance, governed assistant

evaluation discipline

A structured practice for checking AI outputs against quality, safety, business, and compliance expectations before and after launch.

Also referenced as: AI evaluation, eval discipline

human review

A deliberate point where an accountable person checks context, risk, quality, or next action before AI-assisted work is accepted or acted on.

Also referenced as: human checkpoint, human decision checkpoint

risk tiering

A way to classify AI use cases by business impact, data sensitivity, regulatory exposure, user population, and required oversight.

Also referenced as: risk tiers, risk classification

Operating Model

AI operating model

The repeatable management system for selecting AI use cases, assigning owners, governing risk, evaluating outputs, and moving work from experiment to production.

Also referenced as: AI operating discipline, operating model

Operations

cost visibility

The ability to see AI usage, unit economics, and workflow-level value clearly enough to manage scale rather than only track experimentation.

Also referenced as: spend visibility

deployment readiness

The condition in which a proposed AI workflow has owners, data permissions, tests, review steps, support paths, and risk controls ready for real use.

Also referenced as: production readiness

workflow governance

The practice of governing AI at the level of real work: inputs, tools, decisions, owners, metrics, exceptions, and review loops.

Also referenced as: workflow-level governance, governed workflow

Trust

source provenance

The visible trail showing which source material, assumptions, systems, or documents support an AI-generated answer or recommendation.

Also referenced as: provenance, source trail