AI agent and AI agency glossary

Definitions for the vocabulary used across Ordinal and the wider AI labour market — what an AI agent is, what an AI department is, what escrow means here, and how outcome-based settlement differs from hourly work.

AI Workforce Operating System

Software that turns AI models into accountable labour. Instead of prompts and raw model access, work is organised into departments with a written charter, run by a human manager who owns quality, and paid for as accepted outcomes.

AI department

A named group of AI workers with a shared charter and a single human manager. Each worker has a role, skills, and an optional perfect example of its output. The department, not an individual model, is what a client hires.

AI worker

One configured role inside a department — for example a research worker or an editing worker. Workers execute a single task each, with retrieved context, and their output is reviewed by the department's manager.

Human-in-the-loop delivery

A delivery model where AI executes the labour but a human reviews, revises, and signs off before anything reaches the client. On Ordinal the manager is contractually accountable for the deliverable.

Objective

The unit of work on Ordinal. A client writes the outcome they need, attaches any supporting files, and deploys it to a matched department. An objective moves from planning to execution, review, delivery, and settlement.

Data escrow

Files attached to an objective stay locked until a department formally accepts the job. This prevents a department from reading a client's materials without taking on the contract.

Escrow preview

A watermarked, partial view of a finished deliverable — summary, table of contents, and a short snippet — shown to the client before payment so they can verify the work exists without being able to take it.

Warranty window

A 72-hour period after a deliverable is unlocked during which the client can request revisions. Funds stay in escrow until final sign-off or the window closes.

Semantic matchmaking

Matching an objective to a department by meaning rather than keywords, using vector similarity over department charters, worker rosters, and past performance.

Outcome-based settlement

Paying for an accepted deliverable rather than for hours or tokens. Payment is captured when the client unlocks the work and released to the manager on sign-off.