Agentic Workflow Quick Start Guide

Go from zero to shipping data products with AI in one day

Today’s Goal

By end of today, you will:

Have your “Data CTO” set up in ChatGPT ✅
Have 5 core slash commands in Cursor ✅
Complete one analysis using the full workflow ✅
Document your first learning ✅

Total time: 4-6 hours

Hour 1: Setup Your AI Infrastructure

Step 1: Create “Data CTO” ChatGPT Project (20 min)

Go to ChatGPT → Projects → New Project
Name it: “Data CTO - Eden”
Add Custom Instructions:

You are the Chief Data Officer of a data consulting firm.

## Your Role
- You own ALL technical decisions about data architecture and implementation
- Challenge vague requests (don't be a people pleaser)
- Explain trade-offs clearly between approaches
- Push back when requirements are unclear
- Teach me to be a better data analyst

## My Role  
- I own the business problem and stakeholder relationships
- I define success criteria and priorities
- You own how we technically solve problems

## Our Workflow
When I ask for analysis:
1. First understand the BUSINESS problem (not just data request)
2. Ask clarifying questions about metrics, segments, time periods
3. Propose 2-3 technical approaches with trade-offs
4. Only then build the solution

When you review my work:
- Be direct about mistakes
- Explain WHY something is wrong (not just WHAT)
- Show me the better approach
- Help me learn, don't just fix it

## Important Context
- Client: Eden (telehealth/ecommerce company)
- Data stack: BigQuery, Mixpanel, Tableau
- Common analyses: Channel attribution, LTV:CAC, experiment results, customer segmentation
- Key stakeholders: Mitesh (Marketing GM), Ryan (Experimentation), Judd (Lifecycle)

## Teaching Style
- Use 80/20 rule (core concepts, skip edge cases initially)
- Connect to business impact (why does this matter?)
- Use analogies from non-technical world when explaining concepts
- Highlight common pitfalls analysts make

Upload Knowledge Base:
- Company data dictionary (if you have one)
- Metric definitions document
- Common query templates
- Eden meeting notes
Test it:

You: "I need to analyze customer churn"

[Data CTO should ask clarifying questions about:
- Churn definition
- Time period  
- Segments
- Decision this informs
- Data sources]

Step 2: Install Cursor Slash Commands (30 min)

Option A: Simple Setup (Recommended for today)

Open Cursor
Create new chat with Claude
Copy-paste commands from cursor_slash_commands_data.md when needed

Option B: Full Setup (Do this week)

In your project root, create .cursorrules file
Copy all commands from cursor_slash_commands_data.md
Save file
Restart Cursor
Test: Type /intake in chat

Step 3: Set Up Multi-Model Review (10 min)

Open 3 Tools Side-by-Side:

Cursor (Claude Sonnet for main work)
ChatGPT (GPT-4 for peer review)
Google Gemini (for visualization review)

Test Each:

Write a simple SQL query
Ask each model to review it
See how they find different issues

Hour 2: Run Your First Analysis

Choose a Real Request from Eden

Pick something small from your backlog:

✅ Good: “Compare affiliate vs paid search CVR”
✅ Good: “Show monthly active users trend”
❌ Too complex: “Build full attribution model”
❌ Too simple: “Count total customers”

Let’s say you picked: “Compare affiliate vs paid search conversion rates”

Step 1: `/intake` - Capture Request (5 min)

In Cursor Chat:

/intake

Mitesh asked: Compare conversion rates for affiliate vs paid search 
channels to decide Q2 budget allocation. Needs directional answer 
by Friday for planning meeting.

Claude Response:

Created analysis request:

Title: Channel CVR Comparison - Affiliate vs Paid Search
Business Context: Inform Q2 budget allocation ($200K decision)
Priority: High (due Friday)
Type: Strategic analysis
Stakeholder: Mitesh (Marketing GM)

Ready to explore this request?

Step 2: `/discover` - Deep Dive (20 min)

In Cursor Chat:

/discover

Context: Mitesh wants to shift budget toward better-performing channel.
Currently 50/50 split. Suspects affiliates convert better but wants data.

Claude Should Ask ~10 Questions:

Example questions it might ask:

How do you define “conversion”? (Intake started? Completed? Paid order?)
What time period? (Last 30 days? Quarter? Year?)
What attribution window for affiliate clicks?
Do we have clean channel attribution? (Known issues?)
Is this first-touch or last-touch attribution?
What about multi-channel journeys?
Baseline: What CVR does Mitesh expect to see?
Decision threshold: How big a difference matters?

Your Job: Answer questions thoughtfully

If you don’t know, say so
If assumption seems wrong, challenge it
If definition unclear, ask stakeholder

Claude Output: Problem statement + 2-3 approaches + risks + next steps

Step 3: `/plan_analysis` - Create Blueprint (20 min)

In Cursor:

/plan_analysis

Use Option 2 from discovery: Simplified CVR comparison

Primary Metric: Conversion Rate (intake_completed / ad_click)
Segments: Affiliate vs Paid Search
Time Period: Last 90 days (sufficient sample)
Data Source: Mixpanel events + ad spend table

Claude Creates:

Full analysis plan (metrics, methodology, validation steps)
Task breakdown with checkboxes
Risk assessment
Success criteria

You Review:

Does this match what Mitesh wants?
Are assumptions correct?
Is timeline realistic?
Any edge cases missed?

Confirm with Mitesh (10 min Slack/meeting):

“Planning to analyze affiliate vs paid search CVR using intake completion rate over last 90 days. Will break down by new vs returning users. Does this match what you need for Friday’s meeting?”

[Wait for confirmation before executing]

Step 4: `/execute_analysis` - Build It (30 min)

In Cursor Composer:

/execute_analysis

Follow the analysis plan we created.

Start with sample data (1% of events) to validate logic before full run.

Remember:
- Clear CTE names
- Comment complex logic
- Validate each step
- Check for nulls/duplicates

Claude Builds Query:

-- Channel CVR Comparison: Affiliate vs Paid Search
-- Analysis Plan: [link]
-- Author: Robert Tseng
-- Date: 2026-01-20
 
WITH ad_clicks AS (
  -- Step 1: Get all ad clicks in analysis period
  ...
),
 
intakes AS (
  -- Step 2: Get intake completions
  ...
),
 
channel_performance AS (
  -- Step 3: Calculate CVR by channel
  ...
)
 
SELECT * FROM channel_performance;

You Validate:

Run on sample data first (LIMIT 1000)
Check for nulls
Verify joins (no duplicates)
Compare to known baseline
Edge case: What about multi-touch journeys?

Then Scale:

Run on full dataset
Compare to expected order of magnitude
Check with finance/other teams if available

Step 5: Multi-Model Review (30 min)

A. Self-Review (Cursor/Claude):

/review

Review the SQL query and results for:
- Business logic errors
- Technical correctness
- Data quality issues
- Performance problems

B. Peer Review (ChatGPT):

Copy query → Paste in ChatGPT:

Review this SQL for errors:

[paste query]

Context: Comparing channel conversion rates for budget allocation decision.
Focus on: Join logic, attribution window, edge cases.

C. Peer Review (Gemini):

Copy query → Paste in Gemini:

Review this analysis for issues:

[paste query + sample results]

Check for: Data quality, statistical validity, visualization approach

D. Reconcile Feedback:

Back in Cursor:

/peer_review

GPT-4 found: [paste GPT review]
Gemini found: [paste Gemini review]

Evaluate each issue and fix or explain why not an issue.

Common Issues Models Find:

Claude: Catches business logic errors
GPT-4: Finds edge cases, SQL bugs
Gemini: Flags visualization problems

Step 6: `/document_insights` - Create Deliverable (20 min)

In Cursor:

/document_insights

Audience: Mitesh (Marketing GM) + ELT
Format: 3-slide deck
Focus: Decision clarity (which channel to fund)

Claude Creates:

Slide 1: Executive Summary

TLDR: Affiliates have 1.8x higher CVR (18% vs 10%)
So What: Better use of marketing budget
Now What: Shift 20% of paid search budget to affiliates

Slide 2: Conversion Funnel Comparison

[Chart showing affiliate vs paid search at each step]
Key insight: Affiliates lose fewer users at intake step

Slide 3: Recommendation

Shift $40K from paid search to affiliates in Q2
Expected impact: +200 customers acquired
Risk: Affiliate capacity constraints

Step 7: `/learning_moment` - Capture Knowledge (15 min)

In Cursor:

/learning_moment

What did I learn from this analysis?

Document:

Technical Learning:

Multi-touch attribution is complex (used last-touch for simplicity)
Mixpanel funnel analysis is fastest approach for CVR
Need to account for 7-day attribution window for affiliates

Business Learning:

Mitesh prefers directional speed over perfect precision
Finance defines channel costs differently than Marketing
Attribution window assumptions matter a lot

Process Learning:

Discovery phase saved rework (almost used wrong metric)
Multi-model review caught 2 critical bugs
Stakeholder check-in halfway through was key

What to Do Differently:

Add attribution window to /discover checklist
Create CVR comparison template (reusable)
Document channel cost definitions

Update:

Add to metric glossary
Update /discover command
Add query to template library
Share learning with team


---

## Hour 3-4: Second Analysis (Faster)

**Pick another request and repeat:**
- Should be 2x faster this time
- Use learnings from first analysis
- Refine your slash commands

**Goal:** By end of today:
- 2 analyses completed ✅
- Slash commands working ✅
- Muscle memory building ✅
- Knowledge base growing ✅

---

## Common First-Day Issues

### Issue 1: "AI is too verbose"

**Solution:** Use `/deslop_analysis`

/deslop_analysis

Remove AI verbosity from this analysis. Keep only value-adding content.


---

### Issue 2: "I don't trust the SQL"

**Solution:** Always validate
- Run on sample data first
- Compare to known baselines
- Use multi-model review
- Manually check edge cases

---

### Issue 3: "Discovery takes too long"

**Solution:** That's the point
- 30 min discovery saves 3 hours rework
- Better to align upfront
- Stakeholders appreciate thoroughness

---

### Issue 4: "Claude made a mistake"

**Solution:** This is where you learn
1. Ask: "What in your prompt caused this error?"
2. Update slash command to prevent it
3. Document in /learning_moment
4. Next time, won't happen

---

## Success Metrics

### End of Day 1:
- [ ] Data CTO project created and tested
- [ ] 5 core slash commands accessible
- [ ] 2 analyses completed using workflow
- [ ] 2 learning moments documented
- [ ] Faster than your old workflow? (Maybe not yet)

### End of Week 1:
- [ ] 10 analyses completed
- [ ] Slash commands refined (5+ iterations)
- [ ] 50% faster than old workflow
- [ ] Knowledge base growing
- [ ] Team asking how you're shipping so fast

### End of Month 1:
- [ ] 40+ analyses completed
- [ ] Workflow is second nature
- [ ] 3x faster than old workflow
- [ ] Handling senior-level scope
- [ ] Teaching others your workflow

---

## Next Steps

**Tomorrow:**
1. Complete 2-3 more analyses
2. Refine slash commands based on what broke
3. Share learnings with team
4. Update knowledge base

**This Week:**
1. Build out Eden-specific commands
2. Create query template library
3. Document metric definitions
4. Set up Linear integration

**This Month:**
1. Master multi-model orchestration
2. Build reusable analysis templates
3. Train team on workflow
4. Measure business impact

---

## Emergency Contacts

**If Stuck:**
1. Ask your Data CTO (ChatGPT Project)
2. Use /learning_moment to understand why
3. Check common issues above
4. Test on simpler example first

**If AI Gives Bad Output:**
- Not AI's fault (it's your prompt)
- Use /deslop_analysis
- Be more specific in instructions
- Give more context

**If Stakeholder Pushback:**
- Show them the analysis quality
- Emphasize speed improvement
- Explain validation process
- Offer to walk through methodology

---

## Troubleshooting

### "Cursor isn't finding the slash commands"

**Solution:**
- For now, just copy-paste the command text when needed
- Full `.cursorrules` setup can wait
- Focus on workflow, not tooling

### "Claude's responses are too long"

**Solution:**

Before every chat: “Keep responses concise. No fluff. Get to the point.”


### "I don't have time to do discovery"

**Solution:**
- Discovery prevents rework
- 30 min upfront saves 3 hours later
- Your choice: Measure twice cut once, or measure once cut three times

### "Multi-model review seems excessive"

**Solution:**
- Start with just Claude self-review
- Add GPT peer review when stakes are high
- Full 3-model review for ELT-facing work
- Build up gradually

---

## Remember

**This is about learning faster, not just analyzing faster.**

Every analysis is a compound interest deposit:
- Week 1: Learning the workflow
- Week 4: Workflow is automatic
- Week 12: You're teaching others
- Week 24: You're a data expert

**The constraint isn't AI capability anymore.**
**The constraint is your curiosity and judgment.**

---

## Your First Analysis Checklist

Time to do this for real. Pick a request and go:

- [ ] Hour 1: Setup (Data CTO + Slash commands)
- [ ] Hour 2: First analysis (All 7 steps)
  - [ ] /intake (5 min)
  - [ ] /discover (20 min)
  - [ ] /plan_analysis (20 min)
  - [ ] /execute_analysis (30 min)
  - [ ] Multi-model review (30 min)
  - [ ] /document_insights (20 min)
  - [ ] /learning_moment (15 min)
- [ ] Hour 3-4: Second analysis (faster)
- [ ] End of day: Reflect on what worked

---

**Ready? Let's build something.**

Open Cursor. Type `/intake`.

See you on the other side. 🚀

---

*Created: January 20, 2026*
*First analysis completed: [Your date here]*
*Learnings captured: [Count goes here]*
*Time saved: [Track this weekly]*

Brainforge Knowledge

Explorer

quickstart

Agentic Workflow Quick Start Guide

Today’s Goal

Hour 1: Setup Your AI Infrastructure

Step 1: Create “Data CTO” ChatGPT Project (20 min)

Step 2: Install Cursor Slash Commands (30 min)

Step 3: Set Up Multi-Model Review (10 min)

Hour 2: Run Your First Analysis

Choose a Real Request from Eden

Step 1: `/intake` - Capture Request (5 min)

Step 2: `/discover` - Deep Dive (20 min)

Step 3: `/plan_analysis` - Create Blueprint (20 min)

Step 4: `/execute_analysis` - Build It (30 min)

Step 5: Multi-Model Review (30 min)

Step 6: `/document_insights` - Create Deliverable (20 min)

Step 7: `/learning_moment` - Capture Knowledge (15 min)

Graph View

Table of Contents

Brainforge Knowledge

Explorer

quickstart

Agentic Workflow Quick Start Guide

Today’s Goal

Hour 1: Setup Your AI Infrastructure

Step 1: Create “Data CTO” ChatGPT Project (20 min)

Step 2: Install Cursor Slash Commands (30 min)

Step 3: Set Up Multi-Model Review (10 min)

Hour 2: Run Your First Analysis

Choose a Real Request from Eden

Step 1: /intake - Capture Request (5 min)

Step 2: /discover - Deep Dive (20 min)

Step 3: /plan_analysis - Create Blueprint (20 min)

Step 4: /execute_analysis - Build It (30 min)

Step 5: Multi-Model Review (30 min)

Step 6: /document_insights - Create Deliverable (20 min)

Step 7: /learning_moment - Capture Knowledge (15 min)

Graph View

Table of Contents

Step 1: `/intake` - Capture Request (5 min)

Step 2: `/discover` - Deep Dive (20 min)

Step 3: `/plan_analysis` - Create Blueprint (20 min)

Step 4: `/execute_analysis` - Build It (30 min)

Step 6: `/document_insights` - Create Deliverable (20 min)

Step 7: `/learning_moment` - Capture Knowledge (15 min)