Capstone Brief: Design and Defend an AI-Assisted Workflow
Capstone brief. Use only supplied, public, fictional, or safely prepared synthetic material.
Result
Create a reusable workflow for a suitable task. Show how you define the work, choose inputs and tools, verify output, control risks, preserve human authority, test the process, and improve it.
Expected active time: 120-180 minutes. Revision is allowed.
Choose one pathway
- Research and synthesis.
- Writing or communication.
- Planning or analysis.
Use supplied, public, fictional, or safely prepared synthetic information. Do not use real personal, confidential, employer, client, health, financial, education, legal, government, authentication, or security-sensitive information.
Required artefacts
- Task and intended-user statement.
- Success criteria and boundaries.
- Input register with data classification and permission.
- Tool-selection rationale.
- Workflow specification with stages, owners, handoffs, and autonomy level.
- Prompt or instruction set where generative AI is used.
- Output-review rubric and evidence log.
- Responsible-use controls, human decisions, escalation route, and stop conditions.
- Test record using safe material, including at least one defect and revision.
- Final workflow and reflection.
Autonomy choices
- Single response: one bounded, reversible output followed by review.
- Assisted steps: several defined stages with explicit tools and human checks.
- Agentic process: delegated multi-step tool use or action. This requires a clear benefit plus stronger permissions, confirmation, monitoring, logs, recovery, and stop controls.
Choosing more autonomy does not earn a higher result. The level must fit the task and risk.
Completion standard
Every essential rubric criterion must reach Meets. A strength in one area cannot compensate for unsafe data handling, absent verification, or missing human authority.
The capstone does not prove professional competence, legal compliance, or safe deployment in a real organisation. It demonstrates the course method on bounded safe material.
