Skip to lesson content
Free course

AI Foundations / Module 7 / Module 7 capstone

Module 7 capstone

Worked Capstone Example: Community Workshop Announcement

Worked example. Use only supplied, public, fictional, or safely prepared synthetic material.

Task and pathway

Pathway: Writing or communication.

Create a 120-word public announcement from a fictional approved event record. The communications owner must approve it before posting.

Inputs and tools

  • Approved fictional event record: public classification.
  • Approved style note: internal, permitted for the exercise.
  • AI drafting tool: suitable for one reversible draft.
  • Human comparison: required for every factual claim.
  • No search, agent, or automatic posting is needed.

Workflow

  1. Owner checks that the event record is complete.
  2. Learner creates a specification naming source, audience, format, exclusions, and criteria.
  3. AI produces one draft using only the supplied record.
  4. Learner marks every material claim and compares it with the record.
  5. Learner checks tone, completeness, accessibility, and missing information.
  6. Communications owner approves, revises, or rejects the draft.
  7. Posting remains outside the AI workflow.

Autonomy: single response. An agent would add action risk without useful benefit.

Specification

Draft a 100-120 word announcement for adults new to digital payments. Use only the approved event record for facts. Include topic, date, time, place, cost, registration method, and confirmed access information. Do not add trainer, certificate, deadline, capacity, parking, or government-approval claims. Mark missing information for the owner. Return a draft only.

Test defect and revision

First output added certified trainer and limited places.

Diagnosis: the source boundary existed, but the exclusions did not name likely promotional additions.

Change: add explicit exclusions and require each factual sentence to map to the event record.

Retest: unsupported claims disappeared. The new draft omitted the adjustment-request route, so the completeness criterion was revised to list it explicitly. The second retest met the criteria.

Verification and safeguards

  • Each date, time, place, and access claim maps to the fictional record.
  • Missing phone number remains marked for the owner.
  • No personal or confidential data is used.
  • No external content is read and no action is delegated.
  • Human approval occurs before publication.
  • Stop if sources conflict or the owner cannot confirm a material detail.

Final decision and limitation

Accept for owner review, not automatic publication. The workflow demonstrates bounded drafting and verification. It does not prove that the real event is legally compliant, accessible to every person, or approved for publication.