TRIBE CODING 3.0: PRACTICAL AND TACTICAL: 10X YOUR OPS

Workshop Handout & Implementation Guide

From SOP to Automated System in 4 Hours

Plan → Prompts → Produce → Polish

---

THE 4-PART FRAMEWORK AT A GLANCE

Phase	Time	What You Do	Deliverable
PLAN	30-45 min	Whiteboard workflow, choose tech stack	Workflow diagram + tech decisions
PROMPTS	1-2 hours	Record expert, test manually with examples	Proven prompts (95%+ accuracy)
PRODUCE	2-3 hours	Build automation in code	Working application
POLISH	1-2 hours	Error handling, UX improvements, deployment	Production-ready system

---

PART 1: PLAN (30-45 MINUTES)

The Whiteboard Exercise

Draw this diagram for every process you want to automate:

Questions to Answer

INPUT

• What raw data starts this process?
• What format is it in? (PDF, email, form, spreadsheet, etc.)
• Where does it come from? (upload, email, API, etc.)
• How often does it arrive?

TRANSFORM

• What needs to happen to the data?
• What decisions get made?
• What analysis is required?
• What's the complexity level?

OUTPUT

• What's the final deliverable?
• What format should it be in?
• Who receives it?
• What defines "success"?

VERIFY

• How do we confirm it worked correctly?
• What quality checks are needed?
• Who reviews/approves?
• What's the feedback loop?

Tech Stack Research (15 minutes)

Choose Your AI Engine:

Task Type	Best Model	Why
Long documents (100+ pages)	Claude Sonnet 4.5	200k context window, works reliably, best workhorse right now
Complex reasoning	GPT-4	Strongest logical chains
Code generation	Claude Sonnet 4.5/Opus 4.1	Superior code understanding
Fast/simple tasks	GPT-4.1 mini or Claude Haiku	Cost-effective speed
Vision/image analysis	Grok, Gemini	Best image analysis

PLAN Phase Deliverable Checklist

• ☐ Whiteboard diagram completed
• ☐ INPUT clearly defined
• ☐ TRANSFORM logic mapped
• ☐ OUTPUT format specified
• ☐ VERIFY method determined
• ☐ Tech stack chosen (AI model + tools)
• ☐ Complexity assessment done (simple/medium/complex)

---

PART 2: PROMPTS (1-2 HOURS)

The Critical Phase: Prove It Works Manually First

Step 1: Record the Expert (20-30 minutes)

Setup Required:

• Voice recorder (phone works fine) and our transcribe tool: http://marbles.link/transcribe
• The expert who actually does this task
• 1-2 real examples to work through live
• Quiet space, no interruptions

Questions to Ask While Recording:

1. "Walk me through what you do first"
2. "What are you looking for here?"
3. "How do you decide X vs Y?"
4. "What makes a good output vs bad output?"
5. "What mistakes do beginners make?"
6. "What nuances matter that aren't obvious?"
7. "What's your thinking process at each step?"

Example from Dr. Anthony's Recording:

This becomes the TRAINING DATA for your AI. It's learning to think like the expert.

Step 2: Gather Before/After Examples (15-20 minutes)

What You Need:

• Minimum 3 examples, ideally 5-10
• Each example needs BOTH input AND output
• Real data (redact sensitive info if needed)
• Variety of cases (simple, complex, edge cases)

Example Structure:

Example #	INPUT (What you start with)	OUTPUT (What it becomes)
1	Raw bloodwork panel (3 pages, patient A)	Comprehensive health report (104 pages)
2	Raw bloodwork panel (3 pages, patient B)	Comprehensive health report (97 pages)
3	Raw bloodwork panel (3 pages, patient C)	Comprehensive health report (112 pages)

Step 3: Feed Claude & Let It Design the Prompt (20-30 minutes)

What to Say to Claude:

What Claude Will Return:

A structured prompt with sections like:

• Role Definition: "You are an experienced functional medicine practitioner..."
• Process Methodology: Step-by-step approach
• Decision Criteria: How to choose between options
• Quality Standards: What makes output excellent
• Output Format: Exact structure and sections
• Edge Cases: How to handle unusual situations

Step 4: Test & Refine (30-45 minutes)

Testing Protocol:

1. Run 5-10 new examples through the designed prompt
2. Measure accuracy: Does it match your quality standards?
3. Identify gaps: Where does it miss the mark?
4. Refine the prompt: Add constraints, examples, clarifications
5. Test again until you hit 95%+ consistency

Accuracy Tracker:

Test #	Input Used	Output Quality	Adjustments Made
1	Test case A	85% - missed nuance	Added criteria for X
2	Test case B	92% - formatting off	Specified exact format
3	Test case C	97% - excellent	Minor tweaks

When to Stop Refining:

• ✅ Hitting 95%+ accuracy consistently
• ✅ Output matches your quality standards
• ✅ Edge cases handled appropriately
• ✅ You'd be comfortable shipping this to clients

PROMPTS Phase Deliverable Checklist

• ☐ Expert process recorded (20-30 min audio)
• ☐ 3-10 before/after examples gathered
• ☐ Initial prompt designed by Claude
• ☐ 5-10 tests completed
• ☐ 95%+ accuracy achieved
• ☐ Final prompt documented and saved
• ☐ Ready to automate with confidence

---

PART 3: PRODUCE (2-3 HOURS)

Now We Code (Because We Know Exactly What We're Building)

What You Bring to the Code Builder

Your Input Package:

• ✅ Proven prompts - tested and refined to 95%+ accuracy
• ✅ Example input files - real data you'll be processing
• ✅ Example output files - the finished results you're happy with
• ✅ Whiteboard workflow - your diagram from Part 1
• ✅ Tech stack choices - which LLM, frameworks, tools

What You're Creating:

1. File upload interface - How users give you data
2. Data extraction - Pull relevant info from uploads
3. API integration - Connect to your chosen LLM
4. Prompt injection - Combine your prompt + user data
5. Response handling - Stream or batch process results
6. Output formatting - Generate the final deliverable
7. Delivery mechanism - Download, email, or save

The Build Process: Component by Component

Stage	Time	What You Build	Test Question
1. Upload Interface	20 min	Simple file upload UI, drag-and-drop, file validation	Can I upload a file?
2. Data Extraction	30 min	Parse uploaded file, extract relevant data, structure for prompt	Is data extracted correctly?
3. LLM Integration	30 min	Connect to AI model, inject prompt + data, handle auth	Does API call return response?
4. Response Handling	30 min	Stream or batch response, format output, error handling	Is output matching expectations?
5. Delivery	20 min	Generate final file, download/email, success confirmation	Can user get the result?

Telling Your Code Builder What to Build

Template for Zed.dev + Claude:

Basic App Structure

PRODUCE Phase Deliverable Checklist

• ☐ Upload interface working
• ☐ Data extraction successful
• ☐ LLM API connected and authenticated
• ☐ Prompt + data injection working
• ☐ Response streaming/processing functional
• ☐ Output formatting matches expectations
• ☐ End-to-end test successful
• ☐ Working application (even if ugly)

---

PART 4: POLISH (1-2 HOURS)

From "It Works" to "The Team Can Use It"

These 6 items separate a working prototype from a production-ready system:

1. Loading States (15 minutes)

Problem: Users don't know if anything is happening

Solution:

• Progress bars or spinners
• Status messages ("Uploading file...", "Analyzing data...", "Generating report...")
• Time estimates if possible ("Usually takes 2-3 minutes")

Without this: Users click again, break things, get frustrated

2. Error Handling (20 minutes)

Problem: Things will break, handle it gracefully

Solution:

• Clear error messages (not "Error 500", but "File upload failed. Please check file size is under 10MB")
• Tell users what went wrong and what to do about it
• Fallback options ("Try again" button, "Contact support" link)
• Log errors for debugging

Without this: Cryptic failures, users stuck, support nightmare

3. Input Validation (15 minutes)

Problem: Garbage in = garbage out

Solution:

• File type checks (only accept .pdf, .docx, etc.)
• File size limits
• Data format validation
• Required field checks

Without this: Mysterious failures halfway through processing

4. Output Quality (20 minutes)

Problem: It works but looks amateur

Solution:

• Proper formatting (headings, spacing, structure)
• Professional presentation
• Consistent styling
• Downloadable/shareable format

Without this: Great content, bad first impression

5. User Instructions (15 minutes)

Problem: Can someone else use this without you?

Solution:

• Simple how-to guide or tooltip
• What file to upload
• What to expect
• How long it takes
• Troubleshooting basics

Without this: Only you can run it, defeats automation purpose

6. Success Confirmation (10 minutes)

Problem: Did it work? What happens next?

Solution:

• Clear success message
• Show what was generated
• Download link or next steps
• Option to run again

Without this: User left wondering if it completed

POLISH Phase Deliverable Checklist

• ☐ Loading indicators added
• ☐ Error messages are clear and actionable
• ☐ Input validation prevents bad data
• ☐ Output looks professional
• ☐ Instructions/documentation created
• ☐ Success states clearly communicated
• ☐ Team member can use without help
• ☐ Production-ready deployment

---

REAL-WORLD CASE STUDY: DR. ANTHONY

The Bloodwork Automation That Changed Everything

Metric	Before	After
Time per report	2-3 hours manual analysis	5 minutes upload to result
Weekly capacity	10-15 patients max	50+ patients possible
Report length	5-10 pages basic analysis	100+ pages comprehensive
Consistency	Varies by energy/time	Identical quality every time
Annual hours saved	-	300+ hours
Revenue impact	Limited by time	$150k+ additional capacity
Time investment	-	4 hours total to build

How It Was Built

PLAN Phase (45 minutes):

• INPUT: 3 bloodwork PDF panels per patient
• TRANSFORM: Extract markers → Compare functional ranges → Identify patterns → Generate insights
• OUTPUT: 100+ page comprehensive health report with recommendations
• VERIFY: Doctor reviews 3 sample reports for accuracy
• TECH: Claude Sonnet 4 (long context window for comprehensive reports)

PROMPTS Phase (1.5 hours):

• Recorded Anthony explaining his analysis methodology (25 minutes)
• Gathered 5 before/after examples (real panels + his reports)
• Fed Claude the recording + examples
• Claude designed initial prompt capturing expertise
• Tested on 8 new panels, refined 3 times
• Achieved 97% accuracy (doctor-approved quality)

PRODUCE Phase (2 hours):

• Built upload interface for PDF panels
• Integrated PDF parsing
• Connected to Claude API with proven prompt
• Added response streaming for progress feedback
• Generated formatted output reports
• End-to-end test successful
• Any team member can produce reports - HUGE

POLISH Phase (45 minutes):

• Added progress indicators ("Analyzing cardiovascular markers...")
• Error handling for invalid PDFs
• File validation (must be PDF, under 25MB)
• Professional report formatting
• Simple instructions ("Upload 3 panel PDFs, get comprehensive report")
• Download confirmation and success message

---

ROI CALCULATOR

Calculate Your Automation's Value

Step	Your Numbers
Hours per week on this process currently	_______ hours
Multiply by 52 weeks	_______ annual hours
Multiply by your hourly rate (or employee cost)	$_______ annual cost
Estimated hours after automation	_______ hours/week
Hours saved per week	_______ hours
Annual hours saved (× 52)	_______ hours
Annual value saved (hours × rate)	$_______ SAVED
Time investment to build (4-6 hours)	_______ hours
Your ROI (Annual value ÷ build time)	_______% ROI

---

COMMON OBSTACLES & SOLUTIONS

What Usually Stops People (And How to Beat It)

Obstacle	The Reality	Solution
"I'm Not Technical" The Fear: "I can't code, this won't work for me"	Anthony isn't technical. You're not coding from scratch - you're directing the builder with proven examples. The AI writes the code.	✓ Focus on Parts 1-2 (Plan + Prompts) ✓ Partner with someone technical for Part 3 ✓ Or use no-code tools for simple automations ✓ Your expertise > coding skill
"I Don't Have Time" The Fear: "I'm too busy to invest 4-6 hours"	You'll save 300-500 hours per year. You're too busy NOT to do this. The ROI hits immediately after completion.	✓ Start with highest-pain process ✓ Block 2-hour sessions over 2 weeks ✓ Calculate the cost of NOT automating ✓ First automation pays for itself in days
"My Process Is Too Complex" The Fear: "My work requires too much nuance"	Complex processes are the BEST candidates. That's where the most time is wasted. The recording method captures nuance.	✓ Record the expert (captures nuance) ✓ Show 10+ examples if needed ✓ Keep human review for edge cases ✓ Automate 80%, review 20%

---

NEXT STEPS & RESOURCES

Your 4-Week Implementation Path

Week	Action Items	Deliverable
Week 1 Identify	• List 3 repeatable processes • Calculate ROI for each • Pick highest-value process • Complete PLAN phase	Whiteboard diagram + tech stack chosen
Week 2 Test	• Record expert process • Gather 5-10 examples • Build prompts manually • Test to 95%+ accuracy	Proven prompts that work manually
Week 3 Build	• Set up development environment • Build component by component • Test each stage • Complete end-to-end test	Working automation (even if ugly)
Week 4 Deploy	• Add polish (errors, UX, etc.) • Create documentation • Train team members • Deploy to production • Measure results	Production system + time savings documented

Tools to Install

Tool	Purpose	Setup Time
Zed.dev or VS Code	Code editor with AI integration	10 min
Claude or ChatGPT API access	AI model for automation	15 min
Node.js environment	Backend processing	15 min
Voice recording app	Capture expert knowledge	Built into phone

Support Structure

• Mastermind Slack Channel: Real-time Q&A and troubleshooting
• Weekly Office Hours: Live help with your specific automation
• Code Review Sessions: Get expert feedback on your builds
• Shared Automation Library: Learn from others' successful automations
• Win Sharing: Celebrate results and inspire the community

This Week's Assignment

---

THE CHOICE IS SIMPLE

---

The businesses that automate win.
The businesses that don't become the cautionary tales.

The choice is simple. The time is now. Let's build.

---

---

© 2025+ Make More Marbles | Tribe Coding™ Framework
Questions? Contact: [email protected]