Skip to content

AI 102 · Module 6

Context Canvas: from case to work application

How do you design context for a real AI use case?

Start Section 1
0 / 8 completed
Sync my progressJoin a session to keep progress across devices and share it with your instructor

Info

After you join, your completed lesson numbers are sent to your instructor together with your learner code, so they can see your progress. Use the codes your instructor gave you, and do not enter your name, company secrets or other personal data. Progress is kept with the rest of that session's data.

Learning objectives

  • Name the ten parts of the Context Canvas and the question each one answers
  • Use a procurement example to walk from goal to source, freshness, sensitivity, strategy and failure plan
  • Complete a Context Canvas for your own work scenario and defend it with RAFS and the four strategies
  • Explain when context cannot all be supplied in advance and AI 201 on tools and trusted data becomes relevant

From decisions to design

10 min

Turn your judgment into a Canvas

Over the first five modules you learned to judge which information deserves to be provided, how to provide it, and how a poor design fails. This module brings them together into one deliverable you can take back to work: the Context Canvas.

Core question

How do you design context for a real AI use case?

Deliverables from AI 101 to AI 201
  1. AI 101: AI Use Case, defining the task and how AI may help
  2. AI 102: Context Canvas, deciding what AI needs to know, where it comes from, and how to handle it
  3. AI 201: MCP Tool Canvas, designing how AI safely reaches outside capabilities

Info

The Canvas is not about making the AI smarter. It makes you think through source, freshness, sensitivity and accountability before you use the AI.

Warning

Use illustrative or de-identified content when filling in a Canvas. Do not paste company secrets, client personal data, passwords or unauthorized sensitive information.

Key takeaway: The Context Canvas turns what you have learned into a design others can review and you can take back to work.

10 min

The ten parts of the Context Canvas

The Canvas has ten parts. Each answers one question, and you fill them in order.

  1. 1. Goal
    What work should the AI do?
  2. 2. Instructions
    What requirements, format and limits must the AI follow?
  3. 3. Required Context
    What must the AI know, and why is it needed?
  4. 4. Source
    Where does each piece of information come from?
  5. 5. Authority
    Which source is the Source of Truth (權威資料來源)?
  6. 6. Freshness
    What is its version or date, and how often must it be updated?
  7. 7. Sensitivity
    Is it public, internal, confidential or restricted?
  8. 8. Context Strategy
    For each item choose Include, Retrieve, Remember or Exclude, and write the reason.
  9. 9. Verification
    How will you know the information is correct and still valid?
  10. 10. Failure Plan
    When information is missing, stale, conflicting or unavailable, what do the AI and the user do?

Info

The template also has a Human Review field: which results must a person confirm before they are used. High-impact results should not be used without review.

Key takeaway: The ten parts start at the goal, pass through source and strategy, and end at verification and the failure plan.

Example and template

10 min

Example: a procurement assessment report

An illustrative scenario shows how the Canvas works. All data, suppliers and figures below are illustrative.

Use case

A procurement officer asks an AI to help prepare an assessment report on "office photocopiers" for a manager to review.

Context Canvas for the procurement assessment report (excerpt)
InformationSource and authorityFreshnessSensitivityStrategy and reason
Purchase requestFormal request from the requesting unitThis requestInternalInclude: short, necessary, specific to this task
Company procurement policyProcurement regulations (Source of Truth)Confirm it is the current versionInternalRetrieve: large library, only relevant clauses each time
Supplier quotationSupplier's formal quotationWatch the validity periodInternalInclude: needed for this comparison, small
Past purchasing dataProcurement system recordsUpdated quarterlyInternalRetrieve: large, look up when needed
Manager's report format preferenceStated by the managerUpdate when it changesInternalRemember: used in every later report
Credit card numberNot applicableNot applicableRestrictedExclude: not needed and must not be provided
  • Verification: the officer checks the quotation and policy clauses against the original documents
  • Failure plan: if the current policy cannot be found, the AI says it cannot confirm and does not guess
  • Human review: the final purchase recommendation and budget judgment are a person's decision
Check yourself

In the procurement report Canvas, which set of strategies is most reasonable?

In the procurement report Canvas, which set of strategies is most reasonable?

Key takeaway: Every Canvas row should answer: is the source trustworthy, is it still valid, may it be shared, and which strategy fits.

10 min

The Context Canvas template

Below is a template you can copy. Paste it into your own document, or use the Context Canvas form on this site to fill it in and submit.

Context Canvas template (Markdown)
# Context Canvas

## 1. Goal
- What work should the AI do?

## 2. Instructions
- What requirements must the AI follow?
- What is the output format?
- What are the limits?

## 3. Required Context
| Information | Why needed? |
|---|---|
|  |  |

## 4. Source & Authority
| Information | Source | Source of Truth? | Authority level |
|---|---|---|---|
|  |  |  |  |

## 5. Freshness
| Information | Version / date | Update frequency |
|---|---|---|
|  |  |  |

## 6. Sensitivity
| Information | Public / Internal / Confidential / Restricted |
|---|---|
|  |  |

## 7. Strategy
| Information | Include / Retrieve / Remember / Exclude | Reason |
|---|---|---|
|  |  |  |

## 8. Verification
- How will you confirm the information is correct?
- How will you confirm it is still valid?

## 9. Failure Plan
- Missing:
- Stale:
- Conflicting:
- Unavailable:

## 10. Human Review
- Which results must a person confirm?

Warning

This template is for designing, so you do not need to paste real confidential content. If you must describe sensitive information, write only its category (for example "client list"), never the data itself.

Key takeaway: The template lays the ten questions out as tables. Write categories and reasons; you never need to paste confidential content.

Scenarios across professions

10 min

Six scenarios across professions

The Context Canvas is not tied to one profession. The six scenarios below (A to F) are illustrative. Pick the one closest to your work, or bring your own.

Six scenarios and what to think about
ScenarioWork the AI assists withJudgment focus
A. AdministrationOrganising meeting notes and internal policyRetrieve the current policy version; meeting content may contain personal data
B. TeacherHelping produce class materialsBe clear on material sources and versions; Exclude student data
C. ResearchHelping organise the literatureVerify every citation; watch the sensitivity of unpublished data
D. ManagerHelping with market analysisDate and authority of market data; sensitivity of internal figures
E. ProcurementHelping evaluate suppliersQuotation validity; keep supplier confidential information from leaking
F. StudentHelping complete a research reportCheck the literature yourself; learning and integrity cannot be handed to AI

Note that all six scenarios need a person to review the result and take responsibility. AI can help organise, but citations, figures and the final judgment remain yours.

Check yourself

In Scenario F, a student asks an AI to organise a research report, and the summary cites several papers the student has not read. Which part of the Canvas needs the most attention?

In Scenario F, a student asks an AI to organise a research report, and the summary cites several papers the student has not read. Which part of the Canvas needs the most attention?

Key takeaway: Data differs by profession, but the judgment is the same: source, freshness, sensitivity, strategy and human review.

Practice and review

10 min

Capstone: complete your Context Canvas

The signature deliverable of AI 102 is a Context Canvas. Complete one for an AI task in your own work, and be ready to explain the reason for each choice.

  1. Choose a task
    Pick a task where you already use AI, or would consider using it.
  2. Write the goal and instructions
    State clearly the work to be done, the output format and the limits.
  3. List the information and judge each item
    For each item fill in source, authority, freshness and sensitivity, then choose Include, Retrieve, Remember or Exclude.
  4. Design verification and a failure plan
    Write how you will confirm the data is correct, and what to do when it is missing, stale, conflicting or unavailable.
  5. Mark human review
    Point out which results a person must confirm before they are used.

Warning

Use illustrative, de-identified or public information. Do not enter company secrets, client personal data, passwords, keys or any unauthorized sensitive data into the Canvas.

Key takeaway: The deliverable is a Canvas you can defend: every choice has a reason, and it shows where a person reviews.

10 min

Review your Canvas

When you finish, check your Canvas against the comparison below. A good Canvas does not hold the most information; it has grounds for every item.

A Canvas that needs work

  • Marks everything as Include
  • Lists the source as "the web" or "a colleague" without naming the authoritative one
  • Has no version or date and no update frequency
  • Does not handle sensitive information
  • Has no failure plan and no human review

A more complete Canvas

  • Can explain why each item got its strategy
  • Names the Source of Truth and the order of priority
  • Labels version, date and update frequency
  • Classifies sensitive information and Excludes where needed
  • Has a verification method, a failure plan, and results a person must confirm
  • Relevance: does each item really affect this task?
  • Authority: do you know which source is the Source of Truth?
  • Freshness: do you know how current the data is and how often it needs updating?
  • Sensitivity: is anything included that should not be?
  • Strategy: is there a reason behind each Include, Retrieve, Remember or Exclude?
Check yourself

A classmate's Canvas marks everything as Include and has no failure plan. What should be fixed first?

A classmate's Canvas marks everything as Include and has no failure plan. What should be fixed first?

Key takeaway: A good Canvas has grounds for every item, and includes verification, a failure plan and human review.

Recap and the step to AI 201

10 min

Recap, and the next question

Done

AI 101 teaches you to state the task clearly, AI 102 teaches you to prepare the information correctly, and AI 201 teaches AI to reach outside capabilities safely.

By the end of AI 102, your thinking shifts from "how do I write a better prompt" to "what information does this task need, which sources can I trust, what should be fetched dynamically, and what should not be provided".

From AI 102 to AI 201
  1. If the context the AI needs cannot all be provided in advance, or the information keeps changing
  2. How can the AI obtain it safely?
  3. AI 201: MCP, tools and trusted data

The items marked Retrieve in your Canvas are exactly what AI 201 takes up: what a tool is, what trusted data means, how permissions and sources should be designed, and when AI may fetch data itself. However convenient the tool, a person still reviews the result and stays accountable.

Check yourself

When does your Canvas naturally lead to the questions AI 201 addresses?

When does your Canvas naturally lead to the questions AI 201 addresses?

Key takeaway: First decide what the AI should know. When it cannot all be supplied in advance, learn how to obtain it safely: that is AI 201.

What next?