Capstone: Build and Document Your AI Workflow

The concept

This capstone asks you to do one thing: take one real AI workflow from the personal workflow list you built in the previous lesson, actually run it on a genuine task from your job, and document it in full. "In full" means every piece the rubric checks for is present and verifiable — not a polished summary of what you intended to do, but a record of what you actually did, including the parts that didn't work perfectly on the first try.

The documentation has four required pieces, each pulling directly from a module you've already completed. The prompt — the actual text you used, built using the anatomy from Module 2 (a clear task, context, format, and constraints) and reflecting at least one round of the iteration skill from the same module if your first attempt needed adjustment. The AI's output — the real, unedited result the prompt produced, not a cleaned-up version with the rough edges removed; the rough edges are often the most instructive part. The quality-control check you applied — specifically which of the Module 3 checks you ran (tracing facts to sources, checking the logic, checking the tone, the "comfortable if fact-checked publicly" gut-check) and what it found, including if it found nothing wrong. One lesson learned — something concrete you'd do differently next time, including, where relevant, a privacy, bias, or security consideration from Module 6 that applied to this specific workflow (did it touch sensitive data that needed a placeholder? did it touch a person in a way that warranted a swap-test check? did it involve content you had to treat with the same skepticism as the security lesson's examples?).

The rubric grades on three criteria, and it's worth understanding each one before you submit. Clarity — does it work? The workflow needs to have produced a real, usable output for a real task, not a demo that technically ran but didn't actually help with anything. Safety — is it responsible? The submission needs to show the quality check was genuinely applied (not just mentioned) and that any privacy, bias, or security consideration relevant to the specific task was actually addressed, not skipped because the task "seemed fine." Completeness — are all steps shown? The prompt, the raw output, the check, and the lesson learned all need to be present and legible as four distinct, checkable pieces — not folded together into a vague paragraph that implies all four happened without showing any of them clearly.

Passing this capstone, alongside the six module quizzes, earns the course certificate — a credential meant to represent that you've actually run this process once for real, not just read about it.

Worked example

A customer-support manager submits a capstone built around drafting response templates for the five most common support ticket categories. The prompt: a system-prompt-style instruction (from Module 2) defining the company's tone ("warm, concise, no corporate jargon"), followed by five specific ticket examples with the desired reply pattern for each — a few-shot structure built directly on the Module 2 technique. The output: the five raw drafted templates, including one that came back noticeably too formal on the first attempt — kept in the submission rather than quietly replaced with the fixed version. The quality check: the manager traced every policy claim in the templates (return windows, refund conditions) against the actual current policy document, catching one template that stated an outdated 14-day return window instead of the current 30-day one — a real error the check caught before any template went live. The lesson learned: policy-referencing content needs the fact-check applied every time a template touches a number that can change, not just once when the template is first created — and, as a Module 6 consideration, no real customer names or ticket numbers were used in the prompt, only structurally similar placeholders, per the privacy rule from that module.

This submission passes all three rubric criteria: it worked (five usable templates, one genuinely improved by the process), it was responsible (the fact-check caught a real error, and privacy was handled correctly), and it was complete (all four pieces — prompt, raw output, check, lesson — are distinctly shown, including the imperfect first draft).

Exercise

Choose one task from your personal AI workflow list (previous lesson) and run it for real. Submit your capstone with all four required pieces: the actual prompt you used (showing at least one iteration if your first attempt needed adjustment), the AI's real, unedited output, the specific quality-control check you applied and what it found, and one concrete lesson learned — including any privacy, bias, or security consideration that applied to this specific task.

Before submitting, check your own work against the three rubric criteria: does it genuinely work (a usable real output)? Is it responsible (the check was actually applied, not just mentioned, and relevant Module 6 considerations were addressed)? Is it complete (all four pieces clearly shown, not folded into a vague summary)?

Recap

• The capstone requires four real, verifiable pieces — the actual prompt, the unedited output, the specific quality check applied, and one concrete lesson learned — not a polished summary that implies them.
• It's graded on clarity (does it work), safety (is it responsible, including Module 6 considerations), and completeness (are all four pieces clearly shown).
• Passing this alongside all six module quizzes earns the certificate — proof you've run this process for real, and a workflow you can keep using and extend with the paid "AI for Business Builders" or "Building AI Agents" courses.

Resources

• Anthropic: "Use case guides" — https://docs.anthropic.com/en/docs/about-claude/use-case-guides
• Poynter: "A guide to fact-checking" — https://www.poynter.org/fact-checking/
• Your Module 2 (prompt anatomy), Module 3 (quality control), and Module 6 (privacy, bias, security) lessons in this course — the direct sources for the four required capstone pieces