AI Build Lab / Learn · Agent Workforce · Week 2 · Module 0

Week 2: Make the organization yours

What am I doing this week, and what will my organization do by the end of it?

Staff only · this page and its checks
Source
Current Internal package, pages/anw-w2-m0.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.
In plain English

Last week you met your organization and gave it a first job.

This week you make it yours. You give it your own work, so it stops guessing.

This week is the base for the next two. In Week 3 you build a team of about five agents for one workstream in your business, on top of what you set up now. In Week 4 you run evals, checking how well your agents do their work so you can keep improving it.

What you get

An organization that works from your work, not from general knowledge. You’ll finish with reviewed knowledge in Cairns, an approved procedure for one recurring process, and a fresh task showing how your organization uses it. You can follow its references back to the original sources.

Why it matters

An agent that doesn’t know your work guesses. An agent that has your files can check. If this is your own business, you end up with agents that work the way you actually do things. If you’re setting this up for a client, the same steps bring in their knowledge, so their agents work the way their business does.

How it works

Four modules, in order. Each one uses what the one before it made. You carry the same process through the week.

Your part

You decide what your organization gets to know. You approve what it collects, check what it understood, and decide what is useful enough to keep.

How this week works

This week you set up three separate parts, and your organization uses them together.

  • Your work. Information about your work and how you do it. Start with one recurring process and three completed examples you’re allowed to use. An existing procedure or supporting facts can help, but they aren’t required.
  • Your data sources. Connect where your work lives. In Module 2, Gigawatt shows you which connectors are available in your app, if any, such as Gmail or Microsoft 365, and you choose the one that holds your records. Selected document or email exports work just as well, so you don't need a connector to take part.
  • The record. Connect Langfuse, a tool in your own account that keeps a record of what your agents do in your prepared workbench. It works in Claude Code and in Codex; Lesson 1 of Module 2 shows the one step that differs.

Set up Langfuse before your organization processes any of your material, and check that a test record, called a trace, arrives. Gigawatt handles the technical setup after you approve what will be recorded.

If setup can't be finished, you can still do the rest of the week. Have The Professor note "no recording set up" in your course progress note and carry on. None of the later lessons depends on the recording. One thing to know if you use Windows: tracing hasn't been tested on Windows yet, so setup may stop with an error there. If it does, take the same route: note it, tell your course support, and continue.

This week you only collect those records as you work. You'll use them to inspect and improve the work later.

Your library

The Knowledge Librarian, Sallee, prepares your work for filing in your library, so your organization has the context about your work it needs.

You’ll learn Cairns as you watch that material get processed. Cairns organizes the knowledge into three connected levels:

  • L3: The original. The selected message or document, preserved so you can check it.
  • L2: The cards. L2 is the card catalogue: source cards plus one-fact cards. A source card is a summary with important facts, uncertainties, and references to the original.
  • L1: The index and its waypoints. A reviewed guide that lists your waypoints (one short card per topic, saying which cards to open for a question), the cards' original sources, and the limits of this collection. The answers live in the cards and originals.

The files stay in the existing library/ folder. Cairns is how that knowledge is organized.

You review Sallee’s proposed update before it becomes saved knowledge. Approving the collection of a source and approving what your organization remembers from it are separate decisions.

Coming together: your organization uses all three

Here is the path through the four modules:

  1. Meet more of your organization. See the new agents and learn the six roles you’ll work with this week: Chief of Staff, Sallee, Cipher, Kansa, The Professor, and Gigawatt. Choose one recurring process to carry forward.
  2. Connect and process your work. Set up Langfuse, approve a small source selection, and learn Cairns while your material is processed. See where the originals, source cards, and reviewed index are saved, then hand the evidence to Cipher.
  3. Create and review your procedure. Cipher maps the process for you to confirm, then drafts a standard operating procedure, or SOP. You review the steps, owners, handoffs, and exceptions. Kansa independently checks the draft against the evidence. You decide what needs revision and which version to approve.
  4. Check that it helps. Use the approved SOP in a fresh task. Inspect the result, keep the evidence, and name what you want to improve next.

Chief of Staff coordinates the work. The Professor guides you through the lessons. Gigawatt helps with setup and recovery. Sallee files the approved knowledge.

Each module uses what the one before it made.Every step returns a decision to you.
The Week 2 path in seven steps. In each step your organization does the work on the left and you make the decision on the right: choose a process; approve Langfuse recording; confirm the exact records collected; check and approve Sallee's Cairns update; confirm or correct Cipher's map; approve one exact SOP version after Kansa's review; inspect a fresh job that uses the SOP. Langfuse records the workbench activity throughout, after its verified setup.MODULEYOUR ORGANIZATION DOESYOU DECIDEYour rosterThe six agents you will work withChoose one recurring processand who does the workGigawatt sets up Langfuseand reads back a test traceApprove what is recordedand place your keys privatelyChief previews your selectionGigawatt checks access; a receipt followsConfirm the exact recordsfrom a connection or selected exportsSallee proposes Cairns updatessource cards and index entriesCheck a claim against its originalthen approve, revise, or declineCipher maps what the records showwith its uncertaintiesConfirm or correct the mapCipher drafts; Kansa reviews the SOPChief routes any revisionsApprove one exact versionthen Sallee files itA fresh job uses the approved SOPInspect the resultand name the next need1234Langfuse records the workbench activity throughout, after you approve and verify its setup.

On a small screen, swipe sideways to see the whole diagram.

Your context matters most

Your organization does better when it has relevant information about your work.

You do not need to connect everything. A small permitted sample is enough for this week.

Before an agent collects material, you’ll check what the connection can access, what this particular job will collect, where the material will be saved, and where model processing happens. Connecting an account does not mean importing the whole account.

We’ll use a fictional client-kickoff example to explain the steps. Your actual exercise uses your own permitted evidence.

One additional source of context is your chat history. If you have used ChatGPT or Claude for your work, it holds a lot of how you write, what you’ve worked on and what worked.

Optional: export your ChatGPT or Claude history. This is supplemental, not required for the week. The export may take time to arrive.

When it arrives, save the file somewhere safe, because it can hold everything you have typed. Exporting it doesn't hand it to your organization. You decide later what, if anything, it gets to use.

Already keep your knowledge somewhere?

An Obsidian vault, a notes app, or a Cairns library you built before this course? Keep it. You don’t have to move anything or throw anything out. This week, keep the two separate while you learn how this one works.

Start this week

Use the native aibl-the-professor agent to start Week 2 Module 0. Read course/workforce/week2/prompts/anw-w2-m0.txt and course/workforce/week2/lessons/anw-w2-m0.md. Explain the week, then ask which recurring process I may safely use. Do not save anything. If the named agent is unavailable, pause and ask Gigawatt to check the installation.

Next: Module 1, meet more of your organization.

When you've finished this lesson
Self-assessed. No gate.
AI Build Lab · LearnAgent Workforce, Week 2, welcome