I went back through hundreds of Git commits last week to analyse what I was asking Claude Code six months ago vs what I ask now.
The evolution surprised me when I saw it.
Six months ago: Detailed technical specifications.
Today: Recognisable patterns with verbal flourish.
Here’s what I discovered: You’re not writing better specifications. You’re speaking in patterns the AI’s training data recognises.
Junior Developer: “Build a contact form with validation and email handling”
Senior Developer: “Create a contact form component with name, email, message fields, client-side validation, server-side sanitisation, honeypot spam protection, success/error states, accessible ARIA labels, mobile-responsive layout, and integration with our email service”
Me: “Add a HubSpot Lite lead pipeline”
The third one delivers contact management, activity timeline, deal stages, email integration, task management, conversion reporting, lead scoring, and pipeline velocity metrics.
From four words.
The AI has seen HubSpot implemented thousands of times in its training data. It knows exactly what that pattern looks like. I don’t need to specify features. I need to trigger the pattern.
This isn’t better prompt engineering. It’s pattern recognition fluency.
In This Issue
Why ‘HubSpot Lite’ Beats Detailed Specifications — Pattern recognition delivers more comprehensive solutions than functional requirements
The Don Draper Home Page That Actually Works — Cultural references trigger design patterns the AI recognises instantly
Pattern Recognition vs Prompt Engineering — The profound difference nobody talks about
Content Studio Lite in One Sentence — Social media management without writing the specification
Why The Flourish Unlocks The Pattern — Describing the feeling matters more than listing the features
The HubSpot Lite Discovery
Last month I needed lead tracking for the platform.
What I could have said:
“Build a lead management system with contact database, custom fields, activity logging, deal stages, kanban board, email integration, task management, reminders, conversion tracking, funnel reporting, lead scoring, and pipeline analytics.”
What I actually said:
“Build a HubSpot Lite lead pipeline.”
What Claude Code delivered:
- Complete contact management with custom fields
- Deal stages with drag-and-drop kanban interface
- Activity timeline tracking emails, calls, meetings
- Task management with automated reminders
- Email integration following HubSpot patterns
- Conversion funnel with stage-to-stage metrics
- Lead scoring framework based on engagement
- Pipeline velocity and health monitoring
The AI recognised the HubSpot pattern from thousands of implementations in its training data. It didn’t need me to specify each feature. The pattern reference unlocked the complete implementation.
Four words triggered what would have taken 200 words of specifications - and delivered a more comprehensive result.
This is pattern recognition fluency.
The Don Draper Home Page
The same principle applies to design language.
Two weeks ago, working on positioning for a manufacturing client.
What I said: “Give the home page Don Draper product positioning - make it feel like the confident choice for people who know quality when they see it.”
What Claude Code understood:
- Minimalist design with confident white space
- Bold, declarative headlines (not questions)
- Product-first imagery showing craft and detail
- Testimonials from recognisable brands
- Subtle social proof (years established, clients served)
- CTAs that assume the sale (“Start Your Project” not “Learn More”)
- Sophisticated colour palette avoiding bright urgency
- Copy that sells benefits through confident statements
The AI has processed Mad Men discussions, advertising case studies, classic positioning frameworks. “Don Draper positioning” triggered an entire design and messaging pattern.
I could have written a 500-word design brief. Or I could reference a pattern the AI has seen analysed thousands of times.
The flourish did the work.
Pattern Recognition vs Prompt Engineering
The dangerous myth: “If the AI doesn’t understand, add more detail.”
The reality: More detail on unfamiliar patterns creates MORE problems, not fewer. You’re forcing the AI away from proven implementations into experimental territory where every specification introduces new uncertainty.
Example:
Prompt Engineering: “Create a social media management interface with post scheduling, content calendar, draft management, image upload, multi-platform posting, analytics dashboard, and engagement tracking.”
Pattern Fluency: “Build a Content Studio Lite for social media management.”
The second one triggers scheduling UI patterns, calendar interfaces, multi-platform posting logic, analytics dashboards, and engagement metrics - because the AI has seen Content Studio implementations hundreds of times.
The difference:
Prompt engineering tries to be clearer. Pattern fluency tries to be more recognisable.
Prompt engineering adds detail. Pattern fluency adds context.
Prompt engineering explains what you want. Pattern fluency triggers what the AI has seen working.
Better prompt: Less detail referencing familiar patterns. Worse prompt: More detail forcing unfamiliar approaches.
Why The Flourish Matters
Which raises the question: why does “Lite” matter? Why not just say “HubSpot”?
Because the “Lite” in “HubSpot Lite” isn’t decoration. It’s signal.
It tells the AI:
- Core features, not enterprise complexity
- Proven patterns, not custom innovation
- Fast implementation, not months of development
- Familiar UX, not learning curve
The “Don Draper” in “Don Draper positioning” isn’t fluff. It’s context.
It triggers:
- Confident minimalism over busy urgency
- Product quality over price competition
- Established authority over startup energy
- Sophisticated restraint over aggressive conversion
The flourish is how you tell the AI which variant of the pattern you want.
“Build me a HubSpot” could mean enterprise sales pipeline with forecasting, territories, and complex workflows.
“Build me a HubSpot Lite” means essential CRM with proven UX patterns and rapid implementation.
That one word changed the entire implementation approach.
The Velocity Multiplier
So how does this translate to actual website development?
Here’s how I build production websites in 4 hours while agencies quote 4 weeks.
Not better technical skills. Pattern recognition fluency with boring technology.
When building escudero-auto.com (see Week 42’s £49/month case study), I didn’t write component specifications. I said:
“Build the service showcase section - think Apple product pages but for industrial 3D printing. Six service cards, each highlighting technical capability with confident minimalism.”
What Claude Code delivered:
- Astro component with proven layout patterns
- Tailwind CSS matching Apple’s restrained design language
- Responsive grid (1/2/3 columns) from training data
- Hover effects following established UX patterns
- Mobile-first approach (seen 10,000+ times)
- Accessible markup (ARIA patterns well-documented)
- Performance-optimised (Astro static generation)
One sentence triggered complete implementation. Because “Apple product pages” is a pattern the AI has seen analysed extensively in Astro/Tailwind contexts.
Traditional agency approach:
- Design mockups and client revisions (1 week)
- Write detailed component specifications (4 hours)
- Developer implements from scratch (8-12 hours)
- Debug responsive layout issues (4 hours)
- Accessibility fixes (2 hours)
- Performance optimisation (3 hours)
Pattern fluency with Astro/Tailwind:
- “Apple product page style service cards” (10 seconds)
- AI applies proven Astro/Tailwind patterns (2 minutes)
- Review and content refinement (20 minutes)
This is where the 40x velocity comes from. Astro and Tailwind do the heavy lifting. Pattern fluency just triggers the right implementation instantly.
The Critical Caveat
Before you rush off to try “HubSpot Lite” with your experimental framework…
This only works with the boring tech stack.
Boring, proven technology unlocks AI development velocity. Pattern recognition is the reason why.
When you say “HubSpot Lite” using Django and PostgreSQL - the AI has seen that implementation thousands of times. Proven patterns everywhere.
When you say “HubSpot Lite” using a bleeding-edge framework released six months ago - the AI is guessing. It understands the concept, but it hasn’t seen the proven implementation patterns.
Pattern fluency requires patterns the AI recognises.
That’s why I build with:
- Astro (static sites - pattern-proven since 2021)
- Django (backend - 15+ years of training data)
- Tailwind CSS (styling - seen everywhere)
- PostgreSQL (databases - decades of implementations)
The AI has processed millions of repositories using these technologies. When I reference a pattern, it applies proven implementations instantly.
Try the same pattern fluency with experimental frameworks? You get conceptual understanding but implementation uncertainty.
What about WordPress? The AI has seen millions of WordPress sites. But “HubSpot Lite” in WordPress means: Which plugin? Which theme conflicts? Which version compatibility? Which hosting limitations? Pattern recognition breaks because implementation varies wildly between installations.
With Django/PostgreSQL or Astro/Tailwind - one proven pattern works everywhere. With WordPress - every site is different.
This is where 26 years of experience matters. I’m not discovering new patterns. I’m recognising proven patterns the AI has seen thousands of times, then triggering them efficiently with boring, pattern-CONSISTENT technology.
The AI amplifies pattern recognition. But only when the patterns exist in its training data AND implement consistently.
What This Actually Means
Prompt engineering is real. But it’s not about writing better specifications.
It’s about recognising which patterns from the AI’s training data solve your specific problem, then triggering those patterns with minimal but precise verbal cues.
“HubSpot Lite” works because HubSpot is well-documented, widely implemented, and thoroughly analysed in the AI’s training data.
“Don Draper positioning” works because Mad Men’s advertising philosophy has been discussed and deconstructed extensively.
“Content Studio Lite” works because content management patterns are proven and prevalent.
The flourish isn’t poetry. It’s applied pattern recognition.
Try This Tomorrow
Next time you’re about to write detailed specifications for Claude Code, stop. Ask yourself: “What proven pattern does this resemble?”
Building a contact form? → “Create a HubSpot-style contact form”
Creating a pricing page? → “Build an Apple-style pricing page with confident minimalism”
Adding user authentication? → “Implement Stripe-style authentication patterns”
Building a dashboard? → “Create a Notion-style dashboard interface”
Adding payment processing? → “Build a Stripe-style checkout flow”
One sentence referencing a pattern the AI knows > 10 sentences of specifications.
You’ll know it worked when Claude Code delivers MORE than you asked for.
That’s not over-engineering. That’s the AI recognising the pattern and applying proven implementations you didn’t have to specify.
P.S. - Next Week: The tech stack decisions that looked boring six months ago but enabled 40x velocity today. Why I chose Astro over Next.js, Django over microservices, and PostgreSQL over MongoDB. The unglamorous decisions that actually mattered.
P.P.S. - Pattern Recognition: Want to see how this works in practice? Reply with “PATTERNS” and I’ll send you the framework for identifying which recognised patterns solve your specific business problems. Not prompt templates. Pattern recognition methodology.
Tony Cooper
We Build Stores - Where Pattern Recognition Delivers What Detailed Specifications Cannot
tony.cooper@webuildstores.co.uk
07963 242210
This Week: Pattern recognition fluency - why “HubSpot Lite” delivers more than 50 functional requirements