Capstone: Ship & certify
Write the runbook, hand the automation its owner and review cadence, submit the evidence trail, and claim the Certified AI Automation Practitioner credential.
Shipping isn't the moment the workflow turns on — it's the moment the workflow can survive without you watching it. That takes one document and two commitments.
The runbook (one page, forever useful)
- What it does & why: three sentences, plus the weekly minutes it retires. The business case, preserved for whoever asks in a year.
- How it works: the spec (trigger → steps → paths), where the prompt lives, where the changelog lives, where the test suite lives.
- Operating it: the gate — who approves, the timeout policy; the problem folder — who checks, when; the health minute — what 'normal' looks like (typical run counts, typical unsure-lane rate).
- When it breaks: the pre-mortem table's greatest hits — likely failures and their fixes; how to pause the workflow safely (every platform has an off switch; the runbook says when pulling it is right: 'when in doubt, pause — everything queues, nothing is lost').
- Owner & review: a named owner (you, until it's someone else by name), and a quarterly review date — is it still needed, still correct, still worth its subscription? Automations without review dates run until they're wrong; with them, they run until they're retired with honors.
Certification rubric
- Design (25%) — honest scoring and payback math; a spec that marked its AI seats and gates before building; a pre-mortem that named real failures.
- Build quality (30%) — contract prompts with enumerated outputs and uncertainty escapes; validation before branching; asymmetric routing defended by error costs; retry-safety answered in writing.
- Operation (30%) — the two-week soak with logs; success criteria measured, not vibed; problem folder and gate genuinely operated (reviewers check the approval log and the operator log — lived-in beats pristine).
- Handover (15%) — a runbook someone else could operate from; named owner; review date; the pause procedure.
Passing earns the Certified AI Automation Practitioner credential (ID format EDOVA-AUT-2026-XXXX, independently verifiable at edova.ai/verify). It attests the loop this course drilled: find the right chore, build deterministic rails, add AI judgment behind contracts and validation, gate what matters, and operate what you ship.
Three onward paths, by appetite: Prompt Engineering & Context Design to make your contract prompts genuinely production-grade (your test-suite habit is its eval discipline, writ small); AI for Business Analysts if choosing what to automate — the investigation before the workflow — is the muscle you want next; and Agentic AI Systems when a workflow's branches stop being enough and the AI needs to choose its own steps (you'll recognize every guardrail there — they're your gates, grown up). Your first post-course builds are already chosen, though: your opportunity map's #2 and #3 have been waiting since Module 1, and each will take a fraction of the time the capstone did. Your automations, meanwhile, keep running — check the problem folder Monday.