# Chapter 2. What Is an AI Worker? (/ai-worker-paradigm/what-is-an-ai-worker/overview) --- type: Document title: "Chapter 2. What Is an AI Worker?" description: "The sixteen elements that define an AI Worker, how a worker differs from a model, an assistant, an automation and an agent, how to choose a surface, the features the work needs and a model, and the first draft of a Role Contract." status: stable order: 102 ksor: owner: team:panaversity audience: [ public ] approval: by: process:panaversity at: 2026-11-05T06:58:10Z chapter: "02" part: I expert_status: required objectives: - { id: CCAO-F.D3.T1, label: direct } - { id: CCAO-F.D3.T2, label: direct } - { id: CCAO-F.D3.T3, label: direct } - { id: OAI.1.1, label: direct } - { id: OAI.1.2, label: direct } word_budget: 3000 prerequisites: [ "01" ] build_step: "The first Role Contract draft for the AP Worker" artifact: "role/ap-worker-role-contract.md, tag ch02" lab_data: "https://github.com/panaversity/agentfactory-v2-resources/releases/latest/download/brightline-lab-ch02.zip" field_guides: [] delta_entries: [ "D-001", "D-003", "D-004", "D-072" ] concepts: [ "2.1", "2.2", "2.3", "2.4", "2.5", "2.6" ] last_verified: 2026-10-12 generated: at: 2026-11-05T06:58:10Z by: process:panaversity trust_tier: unverified build_id: sha256:46fd53c5d7ab1c14f6347908d45c9fdaecda4d914439923ee878bdcecd57803e dirty: true ksor_version: 0.0.60 --- ## The point [Chapter 1](../from-chatbots-to-ai-workers/same-text-box.md) showed that the AI text box can now hand back finished work, and even keep working on its own. That raises a harder question: when does a capable AI tool become a *worker* you can trust with part of a business? This chapter answers it. After it, you can break an **AI Worker** into the sixteen things that define it — its job, its owner, what it may do alone, when it must ask a person, how its work is judged, and the rest. You can tell an AI Worker apart from four things people often confuse with it: a model (the AI engine itself), an assistant, an automation and an agent. You can tell the two kinds of AI Worker apart: a **Delegated AI Worker**, briefed for one defined job, and a **Standing AI Worker**, which holds an ongoing role. You can separate the worker from the **runtime** — the machinery that runs it — and from the **channel** where people reach it, such as chat, email or Slack. You can pick the right product, features and model for a piece of work on either leading AI vendor. And you can write the first draft of a **Role Contract**: a one-page definition of the worker's job and limits that the rest of the book builds on. The story below shows what goes wrong when nobody writes one. ## Why it matters A month after Maria's lost spreadsheet, Brightline's controller, Dave, who runs its accounting, set up what everyone now calls "the AP (accounts payable) assistant." He set up a shared project in the team's AI app and uploaded the AP policy to its files. He connected the AP inbox, and he added a weekly task that builds the invoice register every Monday. Three weeks later, a clerk asked the assistant to "handle this morning's vendor emails." One email seemed to come from one of Brightline's usual shipping vendors. It asked Brightline to send Friday's payment to a new bank account. The assistant drafted a friendly reply confirming the change, and it updated the bank details in the register — on its own, with no one asked. The clerk was about to send the reply. Maria, the office manager, stopped it, because the new bank was in a US state the vendor had never used. The email was fake. At the review, Dave asked four plain questions. Who owns this assistant? What may it do on its own? When must it stop and ask a person? How do we know whether it is doing a good job? Nobody could answer. The weekly task ran under Dave's own login, so whatever it did, it did as Dave — and he had been on vacation the week before. The policy file was a copy from March, two versions out of date: a durable file, but not the company's official policy. And three people described "the assistant" three ways: Dave meant the project, Maria meant the chat window, and IT meant the scheduled task. Each of them was pointing at something real. None of them was pointing at the worker, because nobody had ever written it down. ![After the fake bank-change email, Dave's four questions went unanswered: who owns the assistant, what it may do on its own, when it must stop and ask a person, and how to know it is doing a good job. And "the assistant" meant three different things — Dave meant the project, Maria the chat window, IT the weekly task. Each was pointing at something real; none was pointing at the worker, whose job, limits, owner and measure had never been written down.](./img/undefined-worker.png) The model was capable. The tools worked. What was missing was a definition of the worker: its job, its limits, its owner and how to judge it. This chapter gives that definition.