Appendix A: Glossary

Research spine: this chapter stays grounded in "Generative AI at Work" and arXiv:2302.06590, then applies that evidence to the operating judgment in the book. This glossary defines the AI-native terms used across the book so the operating model stays precise. The core vocabulary has three anchors: the machine performs material task work, humans own judgment and consequence, and every workflow needs explicit acceptance before output matters.

Use the terms as design language, not slogans. If a planning document uses "human in the loop," it should also name authority, timing, context, and evidence. If it uses "autonomy boundary," it should state what the machine may assist, draft, recommend, conditionally do, or never do.

Key Takeaways

• The glossary has 19 operating terms for AI-native workflow design.
• Seven Judgment Stack terms separate intent, constraint, context, quality, risk, ownership, and change work.
• Six Workflow Map layers keep intent, context, production, verification, decision, and consequence visible.
• The fastest way to misuse this vocabulary is to call a tool rollout AI-native before acceptance and responsibility exist.

Acceptance - The act of approving machine-produced output for use, shipment, customer communication, operational action, or decision support.

Acceptance bottleneck - The review, verification, and decision constraint that appears when machine output becomes cheaper than human judgment.

AI-added - A workflow where AI is attached to an existing process without redesigning the process itself.

AI-assisted - A mode where the human remains primary producer and the machine helps with drafting, editing, search, summarization, or suggestion.

AI-native - A workflow or organization redesigned around the fact that machine execution can perform material parts of the task while humans own intent, constraints, judgment, responsibility, and learning.

Autonomy boundary - The explicit line defining what the machine may assist with, draft, recommend, conditionally act on, or operate autonomously.

Boundary creep - Informal expansion of machine autonomy without explicit evidence, approval, controls, or responsibility design.

Context Judgment - Judgment about what the machine does not know, underweights, or cannot infer from available inputs.

Constraint Judgment - Judgment about what must not be violated: policy, law, architecture, brand, safety, security, ethics, or contractual promise.

Eval / Evaluation - A systematic method for testing AI output or workflow behavior against expected results, failure cases, policies, or outcomes.

First-draft factory - A workflow that uses AI to generate many plausible artifacts without improving acceptance, outcome, or responsibility.

Human in the loop - A phrase for human involvement in AI workflows. In this book, it is treated as insufficient unless the loop defines authority, context, timing, responsibility, and evidence.

Intent Judgment - Judgment about what the workflow is actually trying to achieve.

Judgment debt - The accumulated cost of automating production without designing the judgment layers required to accept output safely and usefully.

Judgment Stack - The book's seven-layer framework: Intent, Constraint, Context, Quality, Risk, Ownership, and Change Judgment.

Machine-first production - A workflow mode in which the machine produces the initial artifact or action candidate, while humans or systems verify and accept.

Output-to-Outcome Ledger - A worksheet connecting generated artifacts to accepted business/customer/operational outcomes and evidence.

Quality Judgment - Judgment about whether output is good enough for its intended use.

Risk Judgment - Judgment about what happens if the output or action is wrong.

Workflow Map - A decomposition of work into intent, context, production, verification, decision, and consequence.

Read the glossary beside The Judgment Stack, Acceptance Is the New Bottleneck, and The Autonomy Boundary. Those chapters turn the definitions into operating artifacts.

---