# 2.2 Five things people call "AI" (/ai-worker-paradigm/what-is-an-ai-worker/five-things-called-ai)

---
type: Document
title: "2.2 Five things people call \"AI\""
description: "What a model, an assistant, an automation, an agent and an AI Worker each are, how they nest, and two questions that tell them apart."
status: stable
order: 102.2
ksor:
  owner: team:panaversity
  audience: [ public ]
  approval:
    by: process:panaversity
    at: 2026-11-05T06:58:10Z
chapter: "02"
part: I
expert_status: required
concepts: [ "2.2" ]
last_verified: 2026-10-03
generated:
  at: 2026-11-05T06:58:10Z
  by: esl-rewrite/1.2.0+ksor.1
trust_tier: unverified
build_id: sha256:46fd53c5d7ab1c14f6347908d45c9fdaecda4d914439923ee878bdcecd57803e
dirty: true
ksor_version: 0.0.60
---

**In everyday life.** You go away for a week and leave things to look
after your house. A timer turns on the porch light at 7 p.m. every night,
the same way each time. That is an automation. A voice assistant answers
whenever someone asks it a question, one request at a time — and the
"brain" inside it that turns the question into an answer is the model. A
plumber you call about a leak works out the steps himself, but only for
that one job, and leaves when it is fixed. That is an agent. A house
sitter looks after the whole house. They decide what each day needs, and
they answer to you for how the week went. That is closest to a worker.

People use one word for five different things. Each one is useful. The AI
Worker is the one built to be managed.

| Thing | What it is | Who decides the steps |
| --- | --- | --- |
| Model | The trained engine that turns input into output. On its own it has no tools and no memory | Nobody. It produces one response |
| Assistant | A product that puts a model behind a chat for whoever is typing | The person, one request at a time |
| Automation | Fixed steps written in advance, run the same way every time | The person who wrote the steps |
| Agent | A model in a loop with tools. It plans, acts, checks and acts again | The agent, inside the task |
| AI Worker | A role, defined by its sixteen elements, with a named owner | The worker, inside its authority |

## How the five fit together

The five do not compete. They fit together. A model sits inside every
assistant and every agent. An AI Worker usually does its work through an
agent. It may also call automations for the steps that should never
change, such as saving each invoice PDF to the right folder. An
automation is still the right tool when the steps are known and judgment
adds only risk.

![How the five fit together. On the left, the AI Worker: a managed role
defined by its sixteen elements, shown by its human owner, bounded
authority and evaluations — the part you write down and keep. Below that,
in a dashed band labeled "does its work through — replaceable," sit an
agent, which chooses the steps within a task in a plan-act-check-repeat
loop with a model inside, and an automation, which follows steps written
in advance. On the right, an assistant: a product that puts a model behind
a chat for whoever is typing; people can reach the worker through it, but
by default no one answers for its outcome. One product can combine several
of these.](img/five-things.png)

*Figure 2.3. The five kinds fit together. A model sits inside every assistant and agent. A worker works through an agent and may call automations — and the machinery underneath is replaceable.*

## Two questions that tell them apart

Two questions tell the five apart. The first is *who decides the steps?*
The table answers it for each one. The second is *who answers for the
outcome?* This question separates a worker from all the others. Ask both
questions about how something is used, not about the product, because one
product can be used in several ways. At Brightline, the clerk who happened
to be typing used the assistant, but nobody answered for what it did.
Brightline used an assistant as if it were a worker. An owner alone is
still not enough. Someone in IT owns the script that saves invoice PDFs,
and that script is still an automation. A worker needs the whole set:
defined responsibilities, bounded authority, continuity (it keeps doing
its work over time), evaluations and an owner who answers for its
results. The sixteen elements in 2.1 spell out that set in full.

The same two questions also sharpen the two kinds of worker from 2.1. A
**Delegated AI Worker** decides the steps, but not when to start: you
brief it, it does the job, and it stops — like the plumber, but with an
owner who answers for the result. A **Standing AI Worker** also decides
*when* work is needed, within its authority: it notices the new vendor
email or the Monday deadline and acts, the way the house sitter does not
wait to be told the plants need water. That initiative is what makes
standing work powerful, and what makes it need the most governance.
[Chapter 9](../../managing-ai-workers/work-that-runs-without-you/governing-initiative.md) teaches how to govern it.