AI terms in plain language
Definitions of the 15 terms that matter in AI transformation — each linked to its primary roadmap stage, with worked examples, decision aids, and failure modes. No hype, no vendor spin, no invented statistics.
The model 5 entries
-
AI transformation
The work of changing how an organization operates so that AI measurably improves real workflows — not simply adopting tools. The five-stage model in one page.
-
AI workflow
A sequence of steps where AI components do work against a defined input and produce a defined output. The unit of AI transformation — and the thing a workflow map documents.
-
Autonomy tier
L0 Suggest through L4 Self-extend — the five levels of decision authority a workflow can be granted, and the guardrails required at each.
-
Executive intelligence
The coordination layer an AI operating model needs at scale — registry, decision log, controller. The artifact the Executize stage builds.
-
Company brain
The emergent organizational intelligence that appears when workflows, agents, and coordination all operate against shared context. Not a purchase — an outcome.
The workflow 4 entries
-
Workflow map
One workflow's data in (start and mid), transforms, integrations, and outputs — the single artifact the Pilot stage produces before any automation runs.
-
Goal-based workflow
A workflow defined by its outcome, not its steps — the agent decides how. The eval set is what makes it safe.
-
Human-in-the-loop
A design commitment that a named human reviews and can override a workflow's output before it takes effect. Three patterns, one metric — the intervention rate.
-
Model evaluation
Measuring a workflow's output against a fixed eval set. Produces the baseline that proves a pilot and the ceiling that makes L3 self-improvement safe.
The technique 4 entries
-
AI agent
A workflow component that takes actions without a human approving each step. Three properties separate agents from chatbots — action-taking, autonomy, and a defined goal.
-
Agent swarm
Multiple specialized agents coordinating to a goal that doesn't reduce to one agent's work. When it's the right design, when it's overengineering.
-
Retrieval-augmented generation
A technique that retrieves relevant documents and grounds the model's response in them. What it solves, what it doesn't, and the silent retrieval failure most pilots miss.
-
Hallucination
A fluent but factually wrong output. The four causes, why most enterprise hallucination is actually retrieval failure, and the design response.
The governance 1 entry
-
AI governance
The mechanisms that control how AI workflows operate — policy, roles, procedural controls, technical controls, and audit. Proportional to a workflow's autonomy tier.
The economics 1 entry
-
Total cost of ownership
The six cost lines that make up an AI workflow's true cost — inference, HITL labor, eval, drift, integration maintenance, retirement. Most ROI models miss three of six.
No entries match this filter yet. More are publishing regularly — subscribe to the weekly brief to know when new entries ship.
Every term names its primary stage
The glossary isn't a dictionary — it's how the roadmap talks about itself. Each entry names its primary stage, with links to the playbook and every related term.