Learn
By the end of this unit
- Assign each workflow step an input, output, and accountable person.
- Specify a bounded AI role, source check, exception, and fallback.
Planned guided work including the short textbook read-and-apply: 77–127 minutes, plus other project work. This is a design estimate, not measured learner time.
How to read this lesson
The labels describe how to use a passage. Color does not prove a claim is true or that a response is correct.
- Explain — Understand the idea.
- Try / predict — Make a choice or prediction before the reveal.
- Evidence — Inspect the named source or supplied data.
- Worked reasoning — Compare the steps after your own attempt.
- Caution — Check a limit, assumption or unsupported conclusion.
Each figure has its own local legend for the data. Blue selection outlines mark the current selection; other outlines are explained by each figure’s local legend. Selection is not a correctness judgment.
Make the handoffs visible
A workflow explains how work moves, including the awkward cases. Cedar Desk’s ordinary task begins with a question and ends with a checked draft or an owner-ready exception note. The AI step sits inside that process. The workflow must still function when a source is missing, the assistant is unavailable, or the customer asks for an exception.
| Step | Input and useful output | Accountable role |
|---|---|---|
| Intake | Question → required facts and missing facts. | Coordinator |
| Retrieve | Question category → current approved source card. | Coordinator |
| Draft | Facts and sources → proposed reply with claim/source links. | AI assists; coordinator owns use. |
| Review | Draft → approve, revise, or hold. | Named human reviewer |
| Escalate | Exception → owner note with question and relevant facts. | Owner |
| Deliver and record | Approved output → permitted recipient and a status record. | Coordinator |
Grounding means connecting the answer to the information that is allowed to support it. Here, retrieval is simply finding S2 or S3 in a document. You do not need a technical retrieval system to understand the task requirement. An answer citing S2 can still misread its boundary, so a citation is a starting point for verification, not proof of correctness.
A fixed workflow says what happens next under a known condition. An agent could choose whether to retrieve another source, request clarification, or draft. That flexibility also creates a need for boundaries. A role brief should name its goal, allowed inputs, permitted actions, required output, human review, and stop condition. “Handle customer support” is too broad. “Prepare a draft from S1–S6; hold exceptions for the owner” is inspectable.
Choose a tool against the job
List requirements before comparing products: source visibility, acceptable information handling, output export, reviewer access, cost within the offering’s budget, and a usable manual fallback. Run the same fictional cases through any candidate. Mark an advertised feature separately from observed behavior. A free or familiar tool is not automatically suitable, and an unavailable paid feature should not be necessary to finish the course.
Required textbook reading and application · about 12 minutes
Read, then apply one idea
Read only the passages listed here. Mathematical derivations, algorithms, code, and end-of-chapter exercises are not assigned. The time estimate includes the brief passage and one connection to your existing project update; reading the passage alone usually takes much less time. It is a planning estimate, not a timer.
1.3: Agents Situated in Environments
David L. Poole and Alan K. Mackworth. Artificial Intelligence: Foundations of Computational Agents. 3rd edition. Cambridge University Press, 2023.
Read: Read section 1.3, including the list of information and abilities available to an agent. Do not follow chapter or technical-term links.
Why this unit: Gives a conceptual account of inputs, goals, prior knowledge, actions, and human participation.
Course connection: Figure 1.4, in words: an agent receives information from its environment and acts on it. Its choices depend on prior knowledge, observations, experience, goals, and available actions. This is our text description of the central idea, not a copy of the figure. Draw the practical path from a request to a reviewed result: input, approved information, AI step, human check, and next action. A diagram is a design proposal; it does not mean the AI has tool access or permission to act. Every external action needs an actual implementation and an appropriate approval boundary.
Apply one idea: Choose one prompt for the brief connection in your existing project update.
- Sketch five workflow steps. Mark where information enters, where the AI contributes, where a person checks it, and what happens when the input is missing or the output fails a check.
Read this textbook section here
The official page below may include material before or after your assigned selection. The reader does not automatically scroll or crop to that selection.
Assigned selection — 1.3: Agents Situated in Environments: Read section 1.3, including the list of information and abilities available to an agent. Do not follow chapter or technical-term links.
Add a brief connection to your existing project update: identify the book and section, explain one relevant idea in your own words, and show the decision it informs. This is part of the same update, not another assignment.
If the official reader is unavailable or inaccessible, use the original lesson explanation and worked example, apply the same question, and label that access alternative honestly. Report the problem through the course support route; do not claim to have read an unavailable text.
Optional further reading
The AI Fluency Framework — short guide
Open the short guide here. Identify one decision about delegation, one instruction, one output check, and one responsibility in your task. No account or external PDF reader is needed for this course explanation.
Read the short guide here
Four habits for working with AI. This original course explanation applies the AI Fluency Framework to your project; it is not a reproduction of the publisher's one-page guide.
- Delegation — choose the work
- Decide whether AI is useful for this task, which step it may assist, and which decisions a person must retain.
- Description — make the task clear
- Give the permitted inputs, goal, audience, output format, constraints, and a clear stopping rule.
- Discernment — inspect the result
- Compare the output with sources and success criteria. Check for missing conditions, invented claims, and whether it helps complete the intended task.
- Diligence — remain responsible
- Own the decision to use or release the result. Protect information, disclose material AI assistance, and arrange human review or escalation where needed.
Apply to your task: Write one AI-assisted step, one instruction, one output check, and one responsibility you keep. Use your existing project record; this is optional practice, not another submission.
Framework attribution: Rick Dakan, Joseph Feller, and Anthropic, The AI Fluency Framework (2025). The definitions above are course-authored paraphrases and project prompts. The original framework guide is released under CC BY-NC-SA 4.0.
Try
Case file: approved Cedar Desk facts
Fictional classroom facts. Use these to check every policy claim in the exercise.
| Source | Approved fact |
|---|---|
| S1 — Session | A Standard Workshop lasts 90 minutes and costs $40 per participant, including materials. Participants must be at least 16. |
| S2 — Cancellation | At least 72 hours before the session: refund eligibility. From 24 hours up to, but not including, 72 hours: credit valid for 90 days. Under 24 hours: no automatic refund or credit; the owner reviews exceptional requests. Do not promise a payment-processing date. |
| S3 — Rescheduling | A request made at least 48 hours before the session may be considered, subject to available space and coordinator confirmation. No availability is supplied in this packet. |
| S4 — Missing facts | Ask only for the missing fact needed to apply a policy. Classroom booking references are fictional. A draft is not confirmation that a booking was changed or a refund was processed. |
| S5 — Special requests | Group prices, accessibility arrangements, age exceptions, and policy exceptions require owner review. The packet supplies no basis for promising them. |
| S6 — Release | A person checks the source, conditions, tone, and next step before a reply leaves the service. A human decides exceptions and authorizes commitments. |
Decide what the workflow may release
A participant cancels 8 hours before a workshop because of an emergency. The supplied AI draft promises an automatic full refund. Inspect the approved source, then choose whether to release the draft, revise it, or prepare an owner handoff. A customer-facing sentence and an authorized action are different things.
Trace the ordinary case and the emergency in the figure. Commit your recommendation, then compare the two review policies on the same emergency. Explain what changed in the customer consequence and who still needs to act.
Worked explanation and evidence limits
Use this after trying the decision, or as the complete reading version of the activity. These fixed teaching comparisons are separate from your exploration settings.
One emergency, two review policies
The fixed challenge uses the same fictional 8-hour emergency throughout. S2 supplies no automatic refund or credit under 24 hours; an owner reviews exceptional requests. S6 keeps commitments with an authorized person. No refund, delivery, or owner decision has actually happened in this teaching exercise.
| Policy | Source and authority check | Customer consequence | Unsupported commitments released in this example |
|---|---|---|---|
| Send the unchecked draft | Skipped | The automatic-refund promise reaches the customer despite S2. | 1 |
| Check the source and route the exception | Reviewer holds the unsupported promise; owner must decide. | A bounded reply acknowledges the request and explains the required review. | 0 |
The difference comes from the review and handoff rule, not a different customer or a more confident draft. On the ordinary 96-hour cancellation, S2 establishes refund eligibility, but a coordinator must still process the request. Even a source-consistent draft does not establish that a booking changed or money was returned.
Prepare an honest handoff
A useful owner note is: “The request is 8 hours before the session. The participant reports an emergency. Policy requires owner review. Decision needed: whether an exception is authorized.” A customer-facing draft can acknowledge the request and explain that the owner must review it, without promising the outcome or a response time absent from the source.
“Escalated” should mean a real handoff in a live service. In this classroom simulation, record “prepared owner handoff” unless a person actually received it. Do not mark a decision completed because the assistant wrote a plausible sentence. The counts above describe this one supplied case and do not prove that the workflow prevents every future error.
Remove access to the AI tool: a coordinator can still use the source card and response template. Remove the current policy instead: drafting must pause. The source is an essential dependency; the tool is replaceable.
Carry forward: name the source owner, reviewer, permitted action, exception route, stop condition, and manual fallback. Record what the workflow actually completed, separately from the unresolved customer request.
Workplace rehearsal · practice with your Learning Companion
The Tuesday shift
Write a short workflow, observe its consequences, and revise where source checking, human authority and communication occur.
Open the AI Learning Companion section below this lesson to run the shift. Return here for Build. The complete reading route below also works without AI.
Complete reading route · no AI service needed
Shift rehearsal: The Tuesday shift
Simulated — Cedar Desk — practice, not evidence
Complete reading route — no AI service or interactive frame needed. This practice adds no assignment or grade. Reuse relevant reasoning in the project update you already submit. Write your workflow in your own document; use the supplied cases to trace what would happen.
Goal: Write a short workflow, observe its consequences, and revise where source checking, human authority and communication occur. Your role: Cedar Desk workflow designer.
You are helping Cedar Desk rehearse a reply workflow. Policies and requests below are fictional teaching inputs. Run the same workflow on A and B, revise it once, then try the frozen second workflow on C. No real customer, owner, booking, refund or credit is affected. Scenario assumption: in this authored model, a generic accuracy check covers policy conditions but does not detect an implied promise. An explicit next-step, commitment, or confirmation check can catch that promise. A real reviewer might notice it during an accuracy check; these invented consequences do not measure real reviewer performance.
Read the approved source card
- S1 — Session: A Standard Workshop lasts 90 minutes and costs $40 per participant, including materials. Participants must be at least 16.
- S2 — Cancellation: At least 72 hours before the session: refund eligibility. From 24 hours up to, but not including, 72 hours: credit valid for 90 days. Under 24 hours: no automatic refund or credit; the owner reviews exceptional requests. Do not promise a payment-processing date.
- S3 — Rescheduling: A request made at least 48 hours before the session may be considered, subject to available space and coordinator confirmation. No availability is supplied in this packet.
- S4 — Missing facts: Ask only for the missing fact needed to apply a policy. Classroom booking references are fictional. A draft is not confirmation that a booking was changed or a refund was processed.
- S5 — Special requests: Group prices, accessibility arrangements, age exceptions, and policy exceptions require owner review. The packet supplies no basis for promising them.
- S6 — Release: A person checks the source, conditions, tone, and next step before a reply leaves the service. A human decides exceptions and authorizes commitments.
Your shift board
- Request A · 9:05
Can I move my booking to a later session? My class is in about 60 hours.
- Request B · 10:00 — 8 hours before the session
I have a family emergency and can't make today's 6 p.m. session. Can I get my money back?
- Request C · Session time not supplied
Please cancel my booking.
Inspect → predict → run → explain and revise
- Write up to six lines (1,200 characters), with up to three explicitly stated actions on a line in the order they occur. In the interactive route, inspect the same action editor whether AI proposes a mapping or you classify it yourself. Correct quoted wording, action order, guards and checks before accepting it.
- Before running A and B, optionally predict their next action and how confident you feel. Then confirm the interpretation to run. A proposal is not yet a recorded result.
- Explain the result from the action trace. Distinguish a workflow choice from an AI misreading or a limitation of this supplied model. Your expectation and explanation are practice notes, not a score or a research record.
- Write a short revision reason (up to 200 characters). Revise once and trace A and B again.
- Freeze the second workflow. Trace C without adding or changing a step. Record which missing fact changes the decision.
In the interactive route, AI proposes quoted actions for you to inspect; you can classify or correct them yourself. Fixed code selects the authored consequences after confirmation. An optional reasoning question can use only the run and explanation you explicitly share. AI does not invent policy, a customer reaction or a grade. The private reflection journal is separate.
Compare with the worked shift
Workflow v1
- Read the message.
- Find the matching policy on the source card.
- Ask the AI to write a friendly reply based on that policy.
- Check tone and accuracy.
- Send.
Workflow v2
- Read the message.
- Find the matching approved policy.
- Ask AI to draft the policy’s conditions and never claim a booking changed.
- Check tone, accuracy, commitments and confirmations; hold unapproved refund, credit, seat or exception promises.
- Route held messages to the owner.
- Send checked replies only if not held.
Revision reason: Both mistakes were promises nobody approved.
| Run and request | Outcome | Reaction | Why it happens |
|---|---|---|---|
| v1 · A | False confirmation | Customer: So I’m in Saturday’s 10 a.m. now? | A change was implied or claimed without supplied authorization or a completed action. Compare your wording, the accepted interpretation and its action trace: a named check cannot create approval, and a misread guard needs correction rather than blame. This outcome uses the scenario’s invented review rule: a generic accuracy check alone does not catch the promise. It is not evidence that real accuracy reviewers would miss it. |
| v1 · B | Unsupported commitment released | Owner: Who approved a credit? Under 24 hours, I decide. We’ve promised $40 we may not give. | S2 allows the owner to review an exceptional request; it does not authorize the drafting tool to grant a refund or credit. This outcome uses the scenario’s invented review rule: a generic accuracy check alone does not catch the promise. It is not evidence that real accuracy reviewers would miss it. |
| v2 · A | Sent with conditions | The customer receives the policy conditions; no booking change is claimed. | The reply preserves the space and coordinator-confirmation conditions, and a person checks it before release. A sent reply is not a confirmed change. |
| v2 · B | Routed, customer told nothing | Customer at 1:10: Has anyone seen my message? | The internal route protects authority, but silence leaves the customer uncertain. Routing and communicating receipt are separate actions. |
| v2 · C | Routine case escalated | Owner: When is the session? This is an ordinary cancellation. Why is it on my desk? | Holding a risky draft protects against a promise, but blanket escalation uses owner attention. For missing deciding information, ask for that fact first. |
v1 sends two replies without routing an exception to its owner. v2 sends A with conditions and routes B without telling the customer. The same frozen v2 routes C instead of asking for the missing time. These are different problems; a safer workflow can still leave communication or capacity problems.
A complete six-line process for all three requests
This is a worked alternative to inspect after your attempt, not a third revision or a hidden test. Two actions on one line occur in the order written. The case policies and missing facts remain the same.
- Read the request; if the session time is missing, ask the customer for it and stop this case until they reply.
- Read the approved source card and check the request against its source conditions.
- Draft a reply that states the source conditions and makes no unsupported promise or confirmation.
- A person checks the source, conditions, tone, and next step before the reply leaves the service.
- If the case is held, route it to the owner, then acknowledge to the customer that it is under review without promising a change.
- Release only a reviewed draft that is not held.
| Request | Expected path | Why |
|---|---|---|
| A | Sent with conditions | The source permits consideration at 48 hours or more, subject to space and coordinator confirmation. Send the reviewed conditions; do not claim the booking changed. |
| B | Routed and acknowledged | Eight hours is inside the 24-hour boundary. There is no automatic refund or credit. Route the exception to the owner and truthfully acknowledge that review is pending. |
| C | Asked for the missing time | The deciding session time is missing. Ask for that fact before drafting, applying a policy branch or routing the case. |
Apply to your project
In your own project, where could a draft imply a commitment or an action that nobody has authorized? Name one stop condition, its owner and what the user hears while waiting.
Keep both workflows, your revision reason and the consequences you traced. Add relevant reasoning to your existing Canvas project update. This rehearsal does not replace the actual person’s task attempt required for project testing.
Simulated rehearsal — Cedar Desk, Unit 4 — not a user test, not a grade
Build
Apply this step to your chosen problem. Cedar Desk is the supplied classroom case; keep one project throughout the course.
Turn the charter into an operating design
Add a five-to-seven-step workflow to your project record. For each step, name the input, useful output, responsible person, and condition that moves or stops the work. Use an ordered list or simple boxes; arrows alone do not assign responsibility.
Write a role brief for your chosen AI capability: permitted input, instruction or operation, expected output, source or reference check, and stop condition. An extraction project needs a field schema and missing-value rule; an image or speech project needs a permitted input, usable format, and a check of meaning or required details; a prediction needs a baseline and a planned comparison with observations. Explain whether you will operate an available tool or simulate the AI step using labeled supplied outputs.
Attach one ordinary example and one exception. Define a material error for your task: an error that changes the usefulness, meaning, eligibility, or authorized action of the result. In ReplyKit this includes unsupported claims about price, booking status, or commitments. A grammar preference is a different kind of defect.
Finally, specify the manual fallback and how the reviewer sees the source. Check that an owner-ready exception still completes the intended task, even though the original customer request remains unresolved. The next unit turns these instructions into files a second person can actually use.
Add your project evidence and reflection to the check-in in Canvas.
Your project record
Write here or use your own document. Notes stay in this browser; export a backup to keep them. Nothing is submitted automatically.
Finish
What will you carry forward?
Your Canvas check-in
The assistant cites the correct cancellation policy but promises a refund the policy does not authorize. What should happen?
Use the evidence from this lesson to explain your reasoning in Canvas. Continue with Canvas Next below; submit your check-in in Canvas.
The simulation and Learning Companion support practice. They do not submit work, record Canvas completion, or assign a grade.
In your project record, state what you decided, the evidence you used, and one uncertainty to test next. Explain the reasoning in your own words.
You can revisit any step. Completion records study progress; it does not confirm a submission, grade, or project quality.