AI Workflow Audit

Long surveys kill completion rates. Ten well-chosen questions extract roughly 80% of what's needed to make strong recommendations, and the skill asks them two at a time in conversation — not as a 50-field intake form. Each question is designed to surface a specific category of automation opportunity: time-consuming weekly tasks, hidden complexity, context-switching costs, synthesis-and-transformation work, predictable communication patterns, repeated decisions, and "we've always done it this way" process debt.

Every task identified gets sorted into four buckets. Replace is where AI output quality is good enough with a quick human spot-check (meeting summaries, boilerplate emails, data formatting). Augment is where the human still does the work but AI handles the grunt (writing first drafts, code review pre-pass, decision-option structuring). Automate is set-and-forget (file organisation, status reports from existing data, commit messages). Skip is honest about what isn't worth automating — anything done less than monthly, anything requiring physical presence, anything that's the core of your professional judgement.

For sensitive or regulated roles — clinical, legal, financial, educational — the skill applies the Admin Surround Principle: validate that the core work is human-only, then audit the supporting tasks around it. A construction manager's site visits stay human; their inspection reports and contractor emails don't. A clinician's diagnoses stay human; their intake summaries and compliance documentation don't. The boundary gets named explicitly so you can hand the audit to a compliance lead without flinching.

The deliverable is a prioritised table — Quick Wins this week, Build This Month, Strategic Automation This Quarter — with estimated hours saved per task and the specific tool to use ("Claude for X, Perplexity for Y", not "use AI"). It ends with a single first move — one concrete action you can take today, timed against the manual version, so you walk away with a working proof of concept rather than a 40-item TODO that you'll never start.