# AI-Native Workflow Checklist
## Design, Build, Test, and Document Practical AI-Assisted Workflows

Use this checklist to evaluate workflows and decide where AI can help—without losing human judgment or control.

---

## PHASE 1: EVALUATE YOUR WORKFLOW

### Workflow Basics
- [ ] **What is the workflow?** (e.g., content creation, lead qualification, proposal writing)
- [ ] **Current owner(s):** Who does this today?
- [ ] **Frequency:** How often does this happen? (daily, weekly, monthly)
- [ ] **Time spent:** How long does each cycle take?
- [ ] **Output quality requirement:** How critical is accuracy/quality?
- [ ] **Decision points:** Where does a human need to make a judgment call?

### Identify Friction
- [ ] **Repetitive steps:** What parts feel mechanical or copy-paste?
- [ ] **Context-switching:** How many tools or documents are involved?
- [ ] **Handoffs:** Are there delays waiting for feedback or approval?
- [ ] **Quality issues:** Are mistakes common? Why?
- [ ] **Scaling pain:** Would this break if volume doubled?

### AI Fit Assessment
- [ ] **Can AI help generate content or first draft?** (Writing, brainstorm, summarize)
- [ ] **Can AI help filter or qualify?** (Scoring, categorizing, ranking)
- [ ] **Can AI help connect dots?** (Pattern recognition, recommendations)
- [ ] **Does AI have reliable access to needed data?** (No data = no AI)
- [ ] **Is the human judgment step clear?** (Where does human override?)

---

## PHASE 2: DESIGN YOUR AI-NATIVE VERSION

### Workflow Redesign
- [ ] **What stays human-driven?** (Judgment calls, creative direction, final approval)
- [ ] **What becomes AI-assisted?** (First draft, brainstorm, data summary)
- [ ] **What becomes fully automated?** (Routing, logging, simple calculations)
- [ ] **New decision points:** Where will AI output require human review?
- [ ] **Fallback plan:** What happens if AI is wrong or down?

### Tool Selection
- [ ] **Which AI tool is best?** (ChatGPT for writing, Claude for analysis, custom for specialized)
- [ ] **How does data flow?** (Paste, API, direct integration, webhook)
- [ ] **Who has access?** (Individual, team, restricted API)
- [ ] **How is prompt consistency maintained?** (Templates, guidelines, system prompts)

### Quality Standards
- [ ] **What does "good enough" look like?** (Define acceptance criteria)
- [ ] **How will we catch mistakes?** (Spot checks, feedback loops, metrics)
- [ ] **How will we measure improvement?** (Time saved, quality change, human feedback)

---

## PHASE 3: BUILD & TEST THE WORKFLOW

### Setup
- [ ] **Create a working document/template** showing the flow
- [ ] **Write the AI prompt(s)** with clear instructions and context
- [ ] **Define input requirements** (What data does AI need?)
- [ ] **Define output format** (Structure, length, tone)
- [ ] **Set up feedback loop** (How does human feedback train the workflow?)

### Testing - First 10 Cycles
- [ ] **Does AI output meet quality bar?** (Test 10 examples, note patterns)
- [ ] **Are decision points clear?** (Can humans confidently say yes/no?)
- [ ] **Are edge cases handled?** (What breaks the workflow?)
- [ ] **Is the prompt accurate?** (Refine based on failures)
- [ ] **Time measurement:** How long does each step actually take?

### Refinement
- [ ] **Improve the prompt** based on failures
- [ ] **Add guardrails** for high-risk outputs
- [ ] **Document exceptions** (Things that need manual intervention)
- [ ] **Create decision rubric** for human approval step
- [ ] **Build in spot-checks** to catch drift over time

### Integration
- [ ] **How does output get to the next step?** (Manual, automated, API)
- [ ] **Who needs to approve/sign off?** (Define approval roles)
- [ ] **How does this connect to downstream systems?** (CRM, Slack, email, etc.)
- [ ] **What metrics will you track?** (Volume, quality, time, cost)

---

## PHASE 4: DOCUMENT & OPERATIONALIZE

### Documentation
- [ ] **Write the SOP:** Step-by-step human instructions
- [ ] **Include the prompt(s):** So others can use or refine
- [ ] **Document decision criteria:** How to judge AI output
- [ ] **Show examples:** Good output vs. bad output
- [ ] **Create troubleshooting guide:** Common failures and fixes

### Team Onboarding
- [ ] **Who needs to learn this?** (Individual, team, leadership)
- [ ] **Create a training video or walkthrough** (Loom is great)
- [ ] **Have team test the workflow** on real examples
- [ ] **Gather feedback:** What's confusing? What works?
- [ ] **Iterate the SOP** based on feedback

### Ongoing Monitoring
- [ ] **Weekly check-in:** Is the AI output still good quality?
- [ ] **Monthly review:** Metrics, feedback, refinements
- [ ] **Quarterly refresh:** Retrain or adjust prompts
- [ ] **Feedback collection:** How is the human using it? Any workarounds?
- [ ] **Cost tracking:** Are we saving time/money as expected?

---

## PHASE 5: SCALE & AUTOMATE FURTHER

### When the Workflow is Stable
- [ ] **Can we eliminate the human approval step?** (High confidence = yes)
- [ ] **Can we automate data flow?** (API, Zapier, Make automation)
- [ ] **Can we route outputs automatically?** (CRM, Slack, email)
- [ ] **Should we add guardrails?** (Duplicate check, quality alerts, caps)
- [ ] **Can we combine multiple AI steps?** (Sequential workflows)

### Documentation for Others
- [ ] **Can someone else run this without support?** (Clear enough?)
- [ ] **Is there a decision tree for exceptions?** (Edge cases documented)
- [ ] **Are prompts versioned?** (Know which version is live)
- [ ] **Is there a rollback plan?** (If AI breaks or drifts)

---

## COMMON AI-NATIVE WORKFLOW PATTERNS

### Pattern 1: First-Draft Generation
**Use case:** Content creation, proposals, emails
- AI generates first draft → Human reviews, edits, approves → Publish/send

### Pattern 2: Qualification & Filtering
**Use case:** Lead scoring, issue categorization, ticket routing
- AI scores/categorizes → Human spot-checks categories → Automatic routing

### Pattern 3: Summarization & Insights
**Use case:** Email summaries, meeting notes, report generation
- AI pulls data → Creates summary → Human reviews for accuracy → Share

### Pattern 4: Data Enrichment
**Use case:** Contact research, market analysis, competitive intelligence
- AI researches/analyzes → Adds to CRM/sheet → Human verification → Decision

### Pattern 5: Brainstorming & Ideation
**Use case:** Content ideas, feature brainstorms, campaign concepts
- AI generates 10 ideas → Human curates top 3 → Team votes → Build top 1

---

## RED FLAGS - WHEN NOT TO USE AI

- [ ] **High-stakes decisions** without human review (Finance, legal, hiring)
- [ ] **Customer-facing without testing** (Quality matters)
- [ ] **When data quality is poor** (Garbage in = garbage out)
- [ ] **When you can't verify the output** (No reliable truth source)
- [ ] **When speed doesn't matter** (Don't add complexity for no reason)

---

## QUESTIONS TO ASK BEFORE DEPLOYING

1. **Is the human judgment step clear?** (Where does AI stop and human starts?)
2. **What's our rollback plan?** (What happens if AI output is wrong?)
3. **Are we measuring the right metrics?** (Time, quality, cost, or something else?)
4. **Is the prompt reproducible?** (Can someone else run this workflow?)
5. **What's the failure mode?** (What's the worst that can happen if AI is wrong?)
6. **How often do we retrain/update the prompt?** (Is there a refresh cadence?)
7. **Is this actually faster than the old way?** (Have we measured honestly?)

---

## TEMPLATE: YOUR AI WORKFLOW

**Workflow Name:** ___________________________________

**Current owner:** ___________________________________

**Current time per cycle:** ___________________________________

**AI tool(s) we're testing:** ___________________________________

**Human judgment step:** ___________________________________

**Success metric:** ___________________________________

**Testing timeline:** ___________________________________

**Rollback plan if it fails:** ___________________________________

---

**Remember: AI is a tool, not a replacement. The workflows that win are the ones where AI handles the repetitive parts and humans own the judgment.**