The Experiment
In September 2025, I started using AI-assisted development for client websites. Not as a novelty — as my primary workflow. Six months and 20+ production sites later, here's an honest breakdown of what worked, what didn't, and what it means for the future.
The Stack
Every site was built with:
- Next.js 15 — React framework with static export
- Tailwind CSS — utility-first styling
- Framer Motion — animations
- Claude Code — AI pair programmer in the terminal
- Cloudflare Pages — hosting and CDN
Total monthly hosting cost across all 20 sites: about $0. Cloudflare's free tier handles everything.
What Changed
Speed
A website that used to take 2-3 weeks now takes 2-3 days. Not because the AI writes perfect code — it doesn't — but because it eliminates the blank page problem. You go from concept to working prototype in hours, not days.
The 80/20 Split
AI handles about 80% of the work:
- Component structure and boilerplate
- Responsive layouts
- Standard animations and interactions
- SEO metadata and structured data
- Repetitive patterns (cards, grids, forms)
I handle the remaining 20%:
- Design decisions and brand alignment
- Copy and messaging strategy
- Performance optimization
- Edge cases and accessibility
- Client-specific business logic
Quality
This surprised me — the quality went *up*, not down. Here's why:
- Consistency. AI doesn't forget to add meta tags to one page but not another
- Best practices. It defaults to semantic HTML, proper heading hierarchy, and accessible markup
- Modern patterns. It uses current React patterns instead of whatever I learned 3 years ago
What Didn't Work
Blind Trust
Early on, I accepted AI-generated code without reviewing it carefully. That led to:
- Unnecessary dependencies bloating the bundle
- Over-engineered solutions for simple problems
- Subtle accessibility issues (missing ARIA labels, poor focus management)
Lesson: AI is a collaborator, not a replacement. Every line still needs human review.
Complex State Management
AI struggles with complex, interconnected state. Forms with conditional logic, multi-step wizards with validation, real-time data sync — these still require manual architecture.
Design Originality
AI produces competent designs, but they can feel generic. The sites that stood out were ones where I provided strong design direction — specific color palettes, reference sites, animation styles — and let the AI execute.
The Process I Settled On
After 20 sites, here's the workflow that works:
1. Discovery (30 min)
- Client call: understand the business, audience, and goals
- Competitive research: screenshot 3-5 competitor sites
- Define: pages, features, and content requirements
2. Design Direction (1 hour)
- Choose a design reference or mood
- Define color system, typography, and spacing
- Sketch the homepage layout (even rough is fine)
3. Build (4-8 hours)
- Set up the project scaffold with AI
- Build page by page, reviewing as I go
- Iterate on copy and layout in real-time
4. Polish (2-4 hours)
- Performance audit (Core Web Vitals)
- Mobile testing on real devices
- SEO checklist (meta tags, structured data, sitemap)
- Accessibility pass
5. Deploy (30 min)
- Push to GitHub
- Connect to Cloudflare Pages
- Configure custom domain and SSL
Total time: 1-2 days for a standard 5-7 page site.
The Business Impact
For Clients
- Faster delivery means lower cost
- Modern tech stack means better performance
- Iterative process means the final product actually matches their vision
For Me
- Higher volume without sacrificing quality
- More time for strategy and client relationships
- Ability to offer competitive pricing
What This Means for Web Development
The traditional model — designer creates mockups, developer codes them over weeks, QA catches bugs, launch after 2 months — is dying. Not because it's bad, but because it's slow.
The new model is:
- Strategy first. Understand the business problem.
- Rapid prototyping. Get something real in front of the client within days.
- Iterate fast. Change direction without sunk cost.
- Ship continuously. Don't wait for "perfect."
AI didn't make me a better developer. It made me a faster one. The taste, judgment, and client empathy — that's still the human part. And it's the part that matters most.
Advice for Getting Started
If you're a developer considering AI-assisted workflows:
- Start with side projects. Build your own site, a friend's site, anything low-stakes.
- Be specific in your prompts. "Build a hero section" produces generic code. "Build a hero section with a gradient overlay, animated text reveal, and a CTA that pulses on hover" produces something usable.
- Review everything. Don't ship code you don't understand.
- Keep learning fundamentals. AI amplifies your skill — if your foundation is weak, it amplifies that too.
The developers who thrive in this era won't be the ones who write the most code. They'll be the ones who make the best decisions about *what* to build and *why*.