CopySAW
How a three-day prototype beat every SaaS vendor and became Amazon's creative engine
The Problem: Headlines at Amazon Scale
Amazon needed thousands of headlines. Monthly. The Everything Store required copy for everything, everywhere—hundreds of campaigns, thousands of placements, global markets.
"Copywriters do have a tool at their disposal and it's a thesaurus and they are terrible. I had somebody tell me once a thesaurus is almost helpful."
While designers wielded Photoshop and developers commanded infinite compute, copywriters faced blank documents with synonym lists. The inequality was crushing talent and constraining output.
Watch the BRXND Talk
Mike Houston shares the full story of building CopySAW at Amazon—from the "almost helpful" thesaurus to shipping 400+ headlines in Q4.
The Failed SaaS Path
Spring 2024. Amazon evaluated commercial AI copywriting tools. Months of demos. Generic outputs. No understanding of Amazon's brand voice. Meanwhile, Q4 deadlines approached.
Alephic proposed something different: Ship code in three days. Test against every vendor. Let results decide.
We didn't pitch. We built.
The Three-Day Build
Direct AI generation
Generic copy
Added brand guidelines
On-brand but boring
What makes people care?
Headlines that felt human
The revelation: Break the task into reasoning steps. First generate insights. Then write from insights.
Result: CopySAW demolished every commercial alternative.
Amazon's Private Token Advantage
CopySAW succeeded because it leveraged Amazon's unique data assets:
Brand Voice
Decades of creative guidelines encoded
Customer Language
Patterns from billions of interactions
Creative Process
How Amazon's best writers actually work
Campaign History
What resonated, what failed, why
Generic tools train on everyone's data. CopySAW trains on Amazon's private tokens—creating intelligence competitors can't replicate.
The AI Triangle: Alephic's Framework for Transformation
AI: Unlimited Generation
Unlimited generation without human constraints.
- Language models provide infinite variation
- Can generate hundreds of options in seconds
- Never experiences creative fatigue
- Operates at the speed of thought
Code: Workflow Orchestration
Workflow orchestration that captures how work actually happens.
- Breaks complex tasks into steps
- Chains reasoning and generation
- Maintains consistency across outputs
- Enables feedback loops and learning
Expertise: Domain Knowledge
Domain knowledge that makes outputs genuinely useful.
- Industry-specific understanding
- Brand voice and guidelines
- Creative best practices
- Quality judgment
How We Built CopySAW: The Triangle in Practice
AI Component - The Generation Engine
- Connected to AWS Bedrock for model access
- Implemented multi-agent architecture with specialized personas
- Each agent tuned for different creative styles (witty, straightforward, poetic)
- Parallel processing for instant variation at scale
Code Component - The Workflow System
- 1
Brief intake and validation
- 2
Generate "reasons to care" (consumer insights)
- 3
Multiple agents create headlines from insights
- 4
Feedback capture and learning loops
- 5
Pattern detection to prevent repetition
Built in Airtable prototype, then AWS Strands for scale. API integrations with Amazon's creative tech stack.
Expertise Component - Amazon's Creative DNA
- Mike Houston's copywriting methodology embedded in prompts
- Brand voice guidelines encoded at every step
- Legal's approved language patterns built in
- Creative director judgment captured in "Taste Agents"
- Continuous learning from approved/rejected headlines
The Synthesis
When combined, these elements created a system that:
- Generates like AI (unlimited options)
- Thinks like Mike Houston (creative expertise)
- Operates like Amazon (integrated workflow)
This wasn't implementing software. It was encoding excellence.
From Prototype to Platform
Production Test
Holiday deadlines. No time for perfect. Amazon ran production on the three-day prototype. Result: 400+ approved headlines, live in paid media.
Enterprise Scale
The prototype evolved into infrastructure: AWS Strands orchestration, multiple AI personas, continuous learning from feedback, full creative workflow integration.
Company-Wide Deployment
CopySAW is now essential creative infrastructure across Amazon: thousands of headlines monthly, multiple teams and divisions, real-time market response.
Owned by Amazon, not rented from a vendor.
The System
Orchestrator Agents
Conducting workflows
Strategy Agents
Generating insights
Copywriter Personas
Distinct voices
Taste Agents
Encoding judgment
Creativity Engine
Preventing repetition
The Impact
The old way
Manual processes constrained by human capacity and tools.
- Copywriters with "almost helpful" thesaurus
- Serial review, slow iteration
- Human capacity bottleneck
The CopySAW way
AI-powered generation at unlimited scale with instant feedback.
- 50+ options in seconds
- Parallel generation, instant iteration
- Unlimited variations in brand voice
"AI can't do the whole job—have AI do half. CopySAW stands for 'copy said another way.' We just try and throw lots and lots of headline options at the copywriter. The human rewrites all of it."
From three-day experiment to essential infrastructure, CopySAW changed how Amazon creates at scale.
Why Custom Beats SaaS
Generic Tools
Off-the-shelf solutions that treat every customer the same.
- One-size-fits-all workflows
- Train on everyone's data
- Rented capabilities
- No competitive advantage
The AI Triangle Approach
Custom systems built on your unique data and processes.
- Your workflows encoded
- Your private tokens leveraged
- Your system owned
- Your advantage compounding
The Strategic Choice
Amazon faced what every enterprise faces:
Option 1: Months evaluating generic tools, implementing compromises, renting capabilities
Option 2: Three days building custom intelligence, encoding excellence, owning the advantage
They chose ownership. Three days to prototype. Three weeks to production. Now essential infrastructure.
The question isn't whether AI can help your creative teams.
The question is whether you want to rent generic tools or own your competitive advantage.
Build your competitive advantage
Want to build custom AI systems that leverage your private tokens? Let's talk about what's possible.