I rebuilt a small SaaS marketing site over a weekend using Cline. The result: a working, polished site shipped in two days, with about 70% of the layout code coming from Cline. The conversion rate post-launch improved 8% (compared to the predecessor site).
This was a high-leverage AI tool project. The work shape — Astro components, Tailwind, mostly content with some interactivity — fit Cline’s strengths.
The project
The site:
- Astro 5 with View Transitions
- Tailwind v4
- Content in MDX
- A handful of interactive components (pricing calculator, feature comparison, contact form)
- Hosted on Vercel
The predecessor was a Next.js site from 2023 that had accumulated cruft. Decision: redesign rather than incremental improvements.
What I prepared
Before opening Cline, I spent a few hours on:
- Wireframes for each page (just rough Figma)
- A list of content sections with copy notes
- A color palette and typography choices
- A reference site I admired (for visual inspiration)
This was the design work. AI doesn’t replace this; it accelerates the implementation.
Saturday: structure and layout
Day 1 work:
- Set up Astro project, Tailwind config, color tokens
- Built the homepage, pricing page, features page, blog index
- About 12 hours of work over the day
Cline handled:
- The Astro layout components
- The Tailwind class compositions
- The View Transitions setup
- The MDX integration
What I did manually:
- Wrote the content (copy, headlines, body text)
- Adjusted the visual details (specific spacings, color choices)
- Decided which sections went where
The pattern: I’d describe a section (“a hero with headline, sub, and CTA buttons; visual on the right”) and Cline would scaffold the Astro component with Tailwind. I’d refine the spacing, colors, and details.
Sunday: interactive components
Day 2:
- Pricing calculator (multi-step form with computation)
- Feature comparison table (with collapsible categories)
- Contact form (with validation and submission)
- Newsletter signup
- Search functionality
About 10 hours of work.
These are more complex. Cline produced first attempts that needed iteration:
- Pricing calculator’s calculation logic was wrong on first attempt; iteration fixed it
- Feature comparison’s collapsing animation was clunky; manual refinement
- Contact form was solid; Cline understood the React Hook Form pattern
Productivity here was lower than Day 1 (around 50% Cline contribution vs 80% the prior day) but still meaningful.
Sunday evening: polish and deploy
The last few hours:
- Mobile responsiveness (mostly fine; some sections needed adjustment)
- Accessibility (Cline missed some labels; I added)
- Performance (Cline’s first attempt at images wasn’t optimized; manual fix)
- SEO (meta tags, structured data; Cline handled most)
- Deploy to Vercel
About 4 hours.
What I’d do differently
A few reflections:
Spend more time on design upfront. I had wireframes but they were rough. Better wireframes would have produced better Cline-generated layouts.
Use the model picker more deliberately. I used Claude 3.5 Sonnet throughout. For some Day 2 work, Sonnet 3.7 (newly released) would have produced better first-attempt output.
Pin the design system file. I had a theme.ts with the color and spacing tokens. Pinning it in chat from the start would have produced more consistent code earlier.
Test mobile earlier. I tested mobile last; had to fix several sections. Earlier mobile testing would have caught issues during initial generation.
What surprised me
A few things:
Cline understood Astro 5 well. I expected gaps in newer framework knowledge. The training data is fresh enough that Astro 5 patterns were correct.
Tailwind v4 was fine. v4 has some syntax differences from v3. Cline picked them up correctly.
View Transitions worked first try. This is a relatively new browser feature. Cline’s setup was correct; the transitions worked on first attempt.
The conversion improvement was real. The new site converts 8% better than the old. Some of this is design (the new site is cleaner); some is performance (it’s faster); some might be psychological (newer site, more recent feel).
Cost
- Cline API spend: $11 over the weekend
- My time: ~26 hours
- Subscription costs: $0 (BYOK)
Total: $11 + my time.
For a marketing site that drives the company’s lead generation, the investment is trivially worth it.
What didn’t fit AI well
Some specific things that Cline couldn’t help with:
Brand voice. The copy needed a specific voice — confident but not pushy, technical but accessible. Cline’s drafts were generic. I rewrote most copy.
Visual judgment. Decisions like “this section should be wider” or “the spacing between these is off” required human eye. Cline’s code matched what I asked for; what I asked for needed my visual taste.
Content strategy. What goes where. Which features lead. What the pricing tiers should look like. These are business decisions Cline can’t make.
Conversion-focused detail. Small things like “this CTA should feel more urgent” required human judgment about user psychology.
For these, AI tools are an aid (Cline could rewrite copy; the rewrite would be generic without my voice). The judgment remained mine.
The honest takeaway
Marketing sites are unusually good AI tool fits because:
- Most of the work is layout and styling (well-trained patterns)
- The interactive components are typically standard patterns
- The visual stakes are bounded (functional > beautiful)
- The deploy is straightforward (Vercel, Netlify, etc.)
For a typical small SaaS marketing site, an experienced engineer with AI tooling can build and ship a polished version in a weekend. This is genuinely faster than pre-AI marketing site work.
For more complex sites (rich animations, novel interactions, brand-distinct visual systems), the AI productivity gain is smaller. The base case is meaningful even so.
Recommendation
For solo founders or small teams considering rebuilding a marketing site:
Start with design. Wireframes and content come from you. AI doesn’t replace this.
Use a fast tooling stack. Astro + Tailwind is well-documented and well-trained. Easier than custom solutions.
Build in a weekend. Time-box. Don’t let perfect be the enemy of shipped.
Deploy frequently. Multiple deploys per day during the build. Each tests the production behavior.
Polish iteratively after launch. Ship, then improve based on real user behavior.
The two-day timeline isn’t aspirational. With AI tooling and clear design, it’s achievable for typical marketing sites. The compound benefit — faster iteration, more frequent improvements — pays off across the site’s lifetime.
