This Website Was Built with AI: Here's How (and Why)

You are reading a website that was entirely conceived, written, designed, and developed with artificial intelligence. Not partially. Not “assisted.” Entirely. From the business brief to production deployment, every step was carried out by a system of 6 specialised AI agents, coordinated by a project manager agent.

This is not a marketing gimmick. It is proof by example that our approach works. And since we believe in transparency, here is exactly how it happened.

Why build a website with AI?

The answer is straightforward: because we are an AI automation consultancy. If we cannot use our own methods to build our own website, why would a client trust us?

But beyond consistency, there is a pragmatic argument. A typical corporate website — brief, design, copywriting, development, SEO — costs between EUR 15,000 and EUR 50,000 and takes 2 to 4 months. We wanted to prove that with AI, you can achieve a professional result in a single day.

Spoiler: it worked.

The multi-agent architecture: 6 specialised AI agents

The secret is not using a single LLM to do everything. It is specialising each agent on a precise domain, with clear instructions, defined constraints, and a structured workflow.

Here are the 6 agents that built piwa.one:

1. Project Manager Agent (@chef-projet-piwa)

Role: Orchestrator. Coordinates the other agents, manages dependencies between deliverables, tracks progress, resolves blockers.

What it produces:

• The project tracking file (docs/SUIVI_PROJET.md) updated in real time
• Architecture and prioritisation decisions
• Validation of each deliverable before moving to the next

2. Business Expert Agent (@expert-metier-piwa)

Role: Interview the founder (me) to extract the substance of the business. Structure the business brief.

What it produces:

• docs/brief-metier.md: PIWA identity, services, differentiation, tone, client cases, vision

Method: Structured interview in 4 blocks (identity, services, clients, vision). The AI asks questions, synthesises answers, requests clarifications. The result is a 3,000-word brief usable by all other agents.

3. Marketing Agent (@expert-marketing-piwa)

Role: Write the complete marketing copy, FR and EN, from the business brief.

What it produces:

• docs/copie-marketing.md: CTA strategy, site architecture, copy for every section (hero, problem, method, services, case studies, founder, meta-transparency, FAQ, contact, footer)

Key detail: French and English versions are cultural adaptations, not translations. The French tone is more direct and punchy. The English tone is more factual and business-oriented.

4. SEO Agent (@expert-seo-piwa)

Role: Optimise copy for organic search and GEO (Generative Engine Optimization).

What it produces:

• docs/strategie-seo.md: target keywords FR/EN, meta tags, JSON-LD, technical recommendations
• Enrichment of marketing copy with naturally integrated keywords

Result: 12 target keywords FR, 12 EN. Structured data (Organization, Service, FAQPage, BreadcrumbList) for each page. Optimised meta descriptions. XML sitemap with cross-language hreflang.

5. Web Designer Agent (@webdesigner-piwa)

Role: Create the complete design system for the site.

What it produces:

• docs/design-system.md: “Dark Precision” concept, colour palette, typography, components, animations, responsive design, image treatment

Chosen concept: A dark, clean, premium site. Near-black background (#0A0A0F), indigo (#4F46E5) and cyan (#06B6D4) accents, Space Grotesk / Inter typography. The “wow” effect comes from the contrast between the restrained background and the vivid accents.

6. Technical Agent (@expert-technique-piwa)

Role: Implement the site in code from all previous deliverables.

What it produces:

• All source code in src/: layouts, components, pages, styles, i18n
• Astro, Tailwind, and Firebase configuration

Stack: Astro (static site generator, zero JS by default), Tailwind CSS, TypeScript, Firebase Hosting.

The workflow: from blank page to live site

The process followed a precise order, with each step feeding the next:

Phase 1 — Business brief (morning) The business expert agent interviewed me. 4 question blocks, follow-ups, requests for clarification. In 45 minutes of exchange, it produced a 3,000-word brief covering PIWA’s identity, 4 service offerings, differentiation (client autonomy), tone, and 3 client cases.

Phase 2 — Design + Copy (in parallel) The web designer agent created the design system while the marketing agent wrote the copy. Both worked from the same brief, ensuring consistency. Combined time: under one hour.

Phase 3 — SEO The SEO agent enriched the marketing copy with target keywords and produced the technical SEO strategy (meta tags, JSON-LD, hreflang, sitemap).

Phase 4 — Implementation (afternoon) The technical agent transformed the 3 deliverables (design system, enriched copy, SEO strategy) into code. Astro build, 15+ pages generated as static HTML, responsive, multilingual, full SEO.

Phase 5 — Review and corrections Critical review, corrections applied (15 issues identified), design fixes (4 issues), final SEO review. Each correction documented in the project tracker.

Total time: 1 day. For a multilingual corporate website (FR + EN), with 15+ pages, complete SEO, a coherent design system, and professional content.

The numbers

Metric	Value
Number of AI agents	6
Pages generated	19+
Languages	2 (FR + EN)
Total production time	1 day
Documentary deliverables	5 (brief, copy, design, SEO, tracking)
Lines of code generated	~4,000
Lighthouse Performance score	95+
Cost of equivalent site (agency)	EUR 15,000 - 50,000
Actual cost (AI infrastructure)	< EUR 15 (1 day of Claude subscription)

What this proves

1. AI does not replace humans. It accelerates them. I was involved at every step: business interview, deliverable validation, arbitration decisions. AI did the execution work. I did the direction work.

2. Agent specialisation is key. A single prompt cannot produce a complete website. Six specialised agents, with precise instructions and clear constraints, produce a far superior result to a generalist agent.

3. Workflow matters as much as tools. The order of steps, the dependencies between deliverables, the format of each document: it is process engineering that determines result quality.

4. Transparency is an advantage. Instead of hiding AI usage, we display it. Because it is exactly what we sell: the ability to use AI to produce professional results, quickly, at lower cost.

The limitations (because transparency means this too)

Not everything was perfect:

• Images: AI does not generate photos. We used Unsplash for visuals and CSS gradients for decorations.
• Testimonials: impossible to generate real client feedback. This section awaits genuine reviews.
• Tone: early copy drafts were sometimes too generic. It took 2 refinement passes to reach the desired “expert/edgy” tone.
• French accents: English-centric LLMs regularly forget French diacritics. A dedicated correction pass was needed.

Why this matters to you

If 6 AI agents can build a professional corporate website in a single day, imagine what they can do for your business processes.

At PIWA, we apply this same multi-agent logic to your automation challenges. Whether it is generating sales proposals, automating document processing, or building an internal AI assistant, the approach is the same: specialise, structure, iterate.

The result? Solutions that work in production, not prototypes gathering dust.

If you want to see what AI can do for your business, start with a 2-hour AI workshop. You will leave with a clear view of opportunities, and proof (this website) that we know how to deliver.

Want to see what AI can do for you? Book a workshop and let us move from theory to practice.