We do not just recommend AI-assisted development to clients. We use it ourselves.
The new carrilagency.com — the site you are reading right now — was designed in Paper and built end to end with Claude Code. No traditional design handoff. No Figma-to-code translation layer. No switching between a dozen tools.
Here is how it worked, what we learned, and why it matters for anyone building a serious website in 2026.
What Paper is, and why we used it
Paper is a design tool built for AI-native workflows. Unlike Figma — which is built for precise, manual design — Paper is built around the idea that AI should do most of the heavy lifting.
We used it to explore the visual direction for the new site: the editorial grid with dashed rail lines, the ticket-stub button aesthetic, the halftone dot pattern in the footer. Paper let us iterate on these ideas quickly, see them rendered in real components, and lock them down before a single line of production code was written.
The result was a tight design system: specific hex values, font pairings (Outfit for headlines, Instrument Sans for body, JetBrains Mono for labels), exact spacing rules, and component behaviour documented clearly enough that Claude Code could execute against it without ambiguity.
What Claude Code is, and how we used it
Claude Code is Anthropic’s CLI tool for AI-assisted software engineering. It runs in your terminal, reads your codebase, and writes, edits, and refactors code the way a senior engineer would — following existing patterns, respecting conventions, and explaining what it is doing.
We used it to build the entire Astro 6 codebase from scratch. That includes:
- 307 pages across three locales (English, Arabic, French) with full hreflang, canonical tags, and RTL support for Arabic
- 212 blog posts translated and structured into a content collection with pagination, category filtering, and locale-aware routing
- 50+ service pages built from a shared template system, with translated content and work cards
- A comprehensive redirect system covering every old Webflow URL from the previous site, resolving redirect chains to their final destinations
The process was genuinely collaborative. We would describe what we needed, Claude Code would write it, we would review and give feedback, and it would iterate. When something was unclear, it would ask. When something was wrong, it would diagnose and fix.
What this is not
This is not a story about replacing developers with AI. It is a story about a small, senior team shipping a complex multilingual website faster and with fewer moving parts.
Claude Code does not make decisions. It does not know what your brand should feel like, what your service positioning should say, or how your navigation should work. Those decisions still required people. What it eliminated was the distance between a decision and its implementation.
When we decided the language switcher should only appear on pages that have actual translations — not on every page — that logic took minutes to implement. When we found a gap in the redirect coverage, the fix was immediate. When a design detail was off by a few pixels, we caught it in the same session and corrected it.
The results
- Migration from Webflow to Astro with zero SEO equity lost (all 301 redirects mapped and configured)
- Three-language site launched simultaneously, not sequentially
- Sub-second load times on a static Astro build with no client-side routing overhead
- A fully translatable content system where adding a new locale later requires minimal effort
What it means if you are building something
The combination of Paper and Claude Code changes the economics of a serious web project. Not because it is cheaper in the obvious sense — it still requires skilled humans with good taste and clear thinking — but because it eliminates the friction between thinking and building.
If you have a clear vision and need a team that works this way, that is what we do. Let us know what you are building.
