AI Documentation Transformation
Using hands-on experimentation to validate AI documentation strategy before betting the company on it.
The Opportunity
Why Build a Crochet Documentation Site?
As documentation leaders navigate the AI transformation, I recognized a critical gap: we're making strategic decisions about AI integration without hands-on experience.
Rather than theorize from the sidelines, I built Crochetly.com as a real-world laboratory to answer fundamental questions:
- Can AI actually write good documentation? What's the quality threshold?
- How do modern doc platforms work? What's changed since traditional CMS systems?
- What does "interactive documentation" really mean? Custom components, not just static pages
- How do you structure content for discoverability? Search, navigation, AI-readability
- What's the operational reality? Deployment, maintenance, scaling
Why Crochet?
Crochet is the perfect test case: complex enough to be interesting (stitches, patterns, techniques), personal enough to maintain authenticity (I actually crochet), and fragmented enough to add value (quality content scattered across ad-heavy blogs). Plus, the community needed it—no comprehensive, clean documentation resource existed.
The Approach
Building in Public: Technology & Architecture
I deliberately chose modern, production-grade tools to mirror real documentation platforms:
Key architectural decisions:
- Astro + Starlight: Modern static site generator optimized for documentation—fast, SEO-friendly, component-based
- MDX over Markdown: Enables custom interactive components within content
- Open-source first: Public GitHub repo to enable community contributions
- Component library: Built reusable elements (stitch library, yarn converter, pattern templates)
- AI-assisted content: Used AI for initial content generation, then refined for accuracy and voice
What I Built
Proving the Concept at Scale
Stitch Library — From 8 to 27
I started with just 8 basic stitches. Using AI-assisted content generation, I scaled the library to 27 stitches and techniques—all while maintaining consistent terminology, formatting, and voice. This became my proof point: AI can accelerate content creation by 5-10x when paired with human expertise.
Interactive Components
Yarn converter: Helping users translate between yarn weights and brands
Pattern templates: Structured formats users can customize
Search and filtering: Built-in search with proper indexing—critical for modern docs
Why it mattered: These weren't just features—they were tangible examples I could bring to team discussions. "Look what I built in a weekend" became a powerful counter to "AI can't do X."
Leadership Application
From Personal Experiment to Organizational Proof
The real value wasn't the site—it was using it as proof-of-concept for organizational change.
The Leadership Moment
In team meetings, I started bringing my actual prompts, GitHub commits, and results. Not presentations about AI—real artifacts. "Here's what I wrote, here's what AI produced, here's what I refined." It shifted the conversation from theory to practice.
Overcoming Resistance
When writers were skeptical about AI, I didn't argue—I showed them. "Here's my commit history. Here's my prompt engineering. Here's the quality difference." Seeing a senior leader get hands-on made it safe for them to experiment too.
Redefining Team Skills
Based on what I'd learned, I reshaped hiring criteria and job requirements:
This project gave me the credibility to make those changes—and the evidence to back them up.
Business Impact
Organizational Outcomes at Cloudflare
Direct applications to my work:
- AI integration strategy: I can make informed recommendations about where AI adds value vs. where human expertise is critical—because I've tested it myself
- Platform decisions: Hands-on experience with Astro/Starlight directly influenced our evaluation of documentation platforms at Cloudflare
- Team adoption: Successfully drove 100% team adoption of AI-assisted workflows by leading by example
- Hiring & skills: Reshaped job requirements organization-wide to include prompt engineering and code literacy
- Velocity gains: Demonstrated 5-10x content creation velocity—validated before rolling out to team
Executive Influence:
- Can speak credibly about AI in documentation from hands-on experience—not just theory
- Demonstrated commitment to staying technical despite executive role
- Shows pattern of experimentation and learning, not just management
- Built tangible proof-of-concept that validates strategic recommendations
Leadership Philosophy: The best leaders don't just direct—they demonstrate. You can't lead teams through AI transformation if you've never written a prompt, shipped a component, or debugged a deployment. This project keeps me credible, current, and effective.