Meet every agent in your organization
Who is in my organization, and how do I find out what each agent can do?
- Source
- Current Internal package, pages/anw-l09.html (Internal b518c35b; src/hunter-week2/SOURCE.md). Original handoff: Internal PR #125.
- Copy
- Exact Internal source package, vendored without editing its lesson or prompt bytes.
- Prompts
- Prompts are rendered from the current vendored Internal package named in Source. Historical publication: student 50ab8c5c from Internal 0e2885b5; that receipt does not qualify later Outlook changes.
- Checked in the app
- Historical baseline QA: bounded integrated checks passed for fresh preparation, upgrade, supported-model native routing and tracing. This receipt does not qualify later Outlook changes. Ordinary-student and Windows qualification remain unverified.
- Open to students
- Yes.
- Other view
- one scroll (Wade chose one part at a time; one scroll is kept to compare)
About this module
These are the agents that will help you turn selected examples of your work into a reviewed SOP.
You’ll meet the expanded roster, learn how to read what an agent does, and follow the six-agent working group you’ll use this week. By the end of the module, you’ll have chosen one recurring process to carry into Module 2, where you choose and collect the records for it.
Keep a possible process in mind as you read. You don’t need to settle on it or gather any material yet.
In Week 1 you worked with some of your agents, one at a time, as a job needed them.
This week your organization has fifteen agents, arranged in six departments. Seven of them are new.
Today you see them together. Gigawatt shows you the list, grouped by department, with the job line that says what each agent does. Then you look at the file that list comes from.
This file matters because of a question you will ask throughout the course: “Why can’t my agent do that?” The agent’s role, tools, and approved connections are good places to start looking. Today you learn to read them.
As you meet the agents, think about one recurring process you might bring to them. Who would help you understand it? Who would check the work? What would still need your decision?
A look at all fifteen agents, which ones you already know and which are new, and a way to check what each agent is set up to do.
“Why can’t it do that?” stops being a mystery. You can check whether the agent lacks the role, the tool or the connection, and fix that one thing instead of rewording your request and hoping.
Gigawatt shows your registry. You read the job lines, choose an agent to ask about, and inspect the record behind the explanation.
You are the one reading. You pick which agent to ask more about and identify one outside source that may hold useful evidence for your process.
Before you start
Open the workbench you prepared in Week 1.
This lesson uses your registry, the file that lists your agents: workforce/shed/registry.yaml. Gigawatt can show it to you inside the conversation.
The list below is what your workbench should have this week. If Gigawatt shows you fewer agents, ask it to check that your workbench is up to date before you continue. A name on this page doesn't prove that agent is installed; Gigawatt's list is what counts.
Start or resume this lesson
Use the native aibl-gigawatt agent to start Week 2 Module 1 Lesson 1. Read course/workforce/week2/prompts/anw-l09.txt and course/workforce/week2/lessons/anw-l09.md. Show one department at a time and wait after each. Do not change the roster. If the named agent is unavailable, pause and ask Gigawatt to check the installation.
See your agents, department by department
Read the job lines, not just the names.
You have worked with some of these already. In Week 1 you called some yourself. The Chief of Staff called others on your behalf, so you may have used an agent without seeing its name.
How you reach a seat. In Claude Code, you @mention a seat at the start of a thread, and that thread stays with it: you can’t switch seats partway through, so start a new thread to reach a different one. In Codex, you reach a seat through the Chief of Staff, which relays your request to it.
Seven are new additions this week. They are marked below. You do not need to remember them all today.
Every agent has a function name, which says what it does. Some also have a nickname, such as Sallee or Cipher. The function name helps you understand the role before you learn the name.
Executive OfficeThree agents: Chief of Staff, Charter Steward, Knowledge Librarian · Sallee
| Agent | What it does | In your organization |
|---|---|---|
| Chief of Staff | Turns your request into a work order (a short written brief of what to do, who does it and what to hand back), routes it to named agents, and brings decisions back to you. | Existing |
| Charter Steward | Checks what your organization says about itself against what is on disk and records your rulings. Reports findings; does not make the decisions. | Existing |
| Knowledge Librarian · Sallee | Prepares source-backed knowledge for your organization’s long-term memory and files approved updates. | Existing |
MarketingOne agent: Voice Analyst · Echo
| Agent | What it does | In your organization |
|---|---|---|
| Voice Analyst · Echo | Reads writing you approve and works out how you sound, quoting the sample behind its claims. Does not draft in your voice. | Existing |
OperationsOne agent: Process Analyst · CipherOne new to you
| Agent | What it does | In your organization |
|---|---|---|
| Process Analyst · Cipher | Maps how one recurring process runs, marks the evidence behind its steps, and prepares an SOP candidate after you confirm the map. | New this week |
QualityTwo agents: Quality Reviewer · Kansa, Evaluation Analyst · EvyOne new to you
| Agent | What it does | In your organization |
|---|---|---|
| Quality Reviewer · Kansa | Independently reviews a draft against its work order, criteria, and evidence. Reports what needs attention; does not fix the draft itself. | Existing |
| Evaluation Analyst · Evy | Grades completed work against criteria you approved. | New this week |
Customer SupportFive agents: Intake Triage Analyst · Cinnamon, Support Lead · Holler, Internal Knowledge Researcher · Hatch, External Researcher · Scratch, Communications Writer · You’ve Got MailFour new to you
| Agent | What it does | In your organization |
|---|---|---|
| Intake Triage Analyst · Cinnamon | Reads an incoming message and identifies what the person is asking for and how urgent it is. | New this week |
| Support Lead · Holler | Plans one customer case and brings together what the researchers find. | New this week |
| Internal Knowledge Researcher · Hatch | Answers a question from your organization’s approved material, with citations. | New this week |
| External Researcher · Scratch | Checks an outside fact for currency using approved sources. | New this week |
| Communications Writer · You’ve Got Mail | Drafts communications for an audience. Does not send, post, or review its own work. | Existing |
FacultyThree agents: Course Companion · The Professor, Operator · Gigawatt, Architect · ArchieOne new to you
| Agent | What it does | In your organization |
|---|---|---|
| Course Companion · The Professor | Explains the course, helps you find your place, and turns a lesson into one next step. Does not do your business work for you. | Existing |
| Operator · Gigawatt | Helps set up the organization, keeps its registry accurate, and explains setup problems and failed runs in plain language. | Existing |
| Architect · Archie | Helps design which agents a job needs and how the work should move between them. | New this week |
Faculty is different from the other departments. Its agents help you learn, design, and set things up. The business agents carry out the work.
Your organization has twelve business agents and three Faculty agents. The wider roster is available to explore. You do not need to find an exercise for every agent this week.
Do this now: ask about the agent that surprised you
Pick a job line you did not expect. Maybe the agent does less than you assumed, maybe more.
Ask Gigawatt, and name the agent you picked:
What does this agent need handed to it before it can work, what does it hand back, and what does it never do?
Keep the answer tied to the installed record. You are learning about the role, not starting a job.
The words you’ll see in your registry
Week 1 introduced many of these. Here they are together because you are about to see them on screen.
| Term | What it means |
|---|---|
| A seat | A role or position: a job your organization is designed to do. |
| An agent | What fills that position and does the work. |
| A department | A group of related seats. |
| The registry | The generated list describing the installed agents, tools, and connectors. |
| A job line | The description of what an agent does. |
A seat is a position. An agent is what fills it.
Your organization has fifteen seats, each with an agent assigned to it.
Here is a shortened teaching example from Echo’s registry record. The comments explain the fields; the job line is shortened for this example. Gigawatt can show your installed record in full.
departments:
- name: Marketing # A department
seats: [aibl-echo] # The seat assigned to that department
seats:
- name: aibl-echo # The agent's identifier
title: Voice Analyst (Echo)
department: Marketing
job: Build an evidence-backed voice package from approved samples. # The job line, shortenedRead it from the top: Marketing contains Echo’s seat. Echo’s record then tells you which agent fills it and what work it does. This example helps you read the registry; it is not something to paste into or edit in your workbench.
The registry describes your agents
Each agent works from its own instructions and available tools. The registry brings information about those agents together so you can inspect it.
- The registry describes the agents. Editing a description is not a reliable way to change what an agent can do.
- Gigawatt rebuilds the registry from the underlying records. A manual edit can be overwritten the next time it is rebuilt.
- If something needs changing, ask Gigawatt to identify the source of that setting and show you a proposal.
Who holds what is recorded in one place. The charter says what your organization may do, and the registry says which agents exist and which tools and connections each one holds. When new agents join or an agent gets a connection, the Charter Steward records the change on your yes. These lists limit each agent, not the main chat you type into. In Claude Code the app holds each agent to its line; in Codex the lines are your records in words, and an agent can reach whatever the chat can.
For this lesson, you only read. You do not need to change an agent or edit the registry.
Later this week, giving the organization your selected source material will help it understand your work. That does not require you to rewrite an agent’s instructions.
Read one agent’s record
Open workforce/shed/registry.yaml in your file editor, or ask Gigawatt to show one agent’s record exactly as written.
Here are the fields you may see:
| Field | What to look for |
|---|---|
title | The agent’s function name and nickname. If you named your Chief of Staff in Week 1, its title shows that name. If you didn’t, the title holds a placeholder until you choose one. |
department | The department it belongs to. |
job | What it does. |
model | Which model it uses. The course runs well on Sonnet, so you don’t need the largest model. |
tools | Which tools it is configured to use. |
skills | The named methods it can load. |
connectors | Its recorded outside-service assignments. |
inputs | What it needs handed to it. |
outputs | What it returns. |
may_read | What it is allowed to read. |
may_write | What it is allowed to write. |
requires_human_approval | Where it must ask you before proceeding. |
never | What it is instructed never to do. |
Some limits are enforced by the tools an agent has. Others are instructions it must follow. Reading a rule is not the same as proving that a particular run followed it.
For example, the Communications Writer does not review its own draft. A separate reviewer checks the work. You will use that same separation when Cipher drafts your SOP and Kansa reviews it.
Do this now: read three lines
Choose one agent and find:
- Its
job. - Its
neverlist. - Its
connectorsentry.
Say one thing you learned from those lines before asking Gigawatt to explain more.
You do not need to understand every field today.
Understand what a connector entry means
Your organization keeps its registry and related lists in workforce/shed/. The course calls this folder the shed. It is a small version of the Tool Shed kept inside your workbench, and you can move it into its own repository later.
It records the agents, tools, and connections the organization is configured to use. The agents themselves live in their own files.
You only need to inspect the registry. Gigawatt handles the supporting programs.
A connector entry might be empty:
connectors: []Or it may list an assignment already made in your workbench.
The connectors line is a label, not a switch. Typing a service name into it does not sign you in or establish access.
There are three separate things to check:
- The account connection: have you signed in and authorized access in the app?
- The agent assignment: which agent is supposed to use that connection, and for what?
- The actual read: can the intended agent read the specific material you approved?
An empty registry entry does not prove that the app has no connected accounts. A filled entry does not prove that a read will succeed.
In Module 2, you will choose a source route, approve a small selection, and check that the intended read works. That connection needs to work for this week’s exercise.
Do this now: name one useful source
Name one place outside the workbench where evidence for your process might live.
“The records for three completed client kickoffs are in Gmail.”
Write it in your own notes. Do not connect the account or collect the material yet.
If you get stuck
| What you notice | What to do |
|---|---|
| An expected Week 2 agent is missing. | Ask Gigawatt to check the installation and explain the difference. |
| Gigawatt’s explanation differs from the installed record. | Ask it to show the record and explain the mismatch. |
| A connector entry is empty. | Leave it unchanged. You will check the connection and assignment during setup. |
| A connector is listed but a read fails. | Ask Gigawatt to distinguish account access, agent assignment, and the attempted read. |
| The registry looks editable. | Leave it unchanged. Ask for a proposal if something is wrong. |
What you’ll keep
Nothing is added to your organization's library in this lesson, and no agent changes. Gigawatt does keep its own short working notes for each job, in the runs/ folder: what it read and what it planned. You don't need to open them.
Keep your note naming one outside source you may want to use.
Before you finish
You do not need to memorize all fifteen agents.
Words you now own
- A seat: a role your organization needs filled.
- An agent: what fills the seat and does the work.
- A department: a group of related seats.
- The registry: the generated list describing the installed agents and their setup.
- A connector assignment: the recorded relationship between an agent and an outside service; it does not sign you in.
Next: Lesson 2, follow this week’s working group and choose your process.