How We Rebuilt the Enersys Website with AI in One Week — A Real-World Case Study

$ git log --oneline --before="2026-02-25" | wc -l
288

$ git log --oneline | grep "initial Next.js"
a1b2c3d feat: ⚡ initial Next.js 14 setup with Turborepo

# commits per day
# Before AI: 288 commits / 540 days = 0.5/day
# After AI: 131 commits / 7 days    = 18.7/day (37x increase)

// article registry pattern — manage all articles from a single place
export const articles: Article[] = [
  {
    slug: 'agentic-ai-enterprise-40-percent-2026',
    title: 'Agentic AI in the enterprise...',
    category: 'ai-tech',
    tags: ['Agentic AI', 'MCP', 'Enterprise AI'],
    readingTime: 7,
    gradient: 'from-[#1e1b4b] to-[#6366f1]',
  },
  // ... 25 more articles
];

$ git log --oneline | head -5
092e6d7 chore: bump version to v1.1.48
09b3e6c style: format think-first command markdown
d75f8d1 style: apply prettier formatting to blog articles
f903ffe fix: fix duplicate opacity attribute in SVG
1e18550 chore: autopublish

$ ls .claude/agents/ | head -12
article-writer.md
blog-publisher.md
code-reviewer.md
content-strategist.md
deploy-manager.md
design-system-architect.md
devops-engineer.md
documentation-writer.md
frontend-developer.md
qa-engineer.md
security-auditor.md
seo-specialist.md

Team Structure

Leadership Layer:

• Tech Lead — architectural decisions, code review standards
• Content Strategist — editorial calendar, content quality

Development Layer:

• Frontend Developer — React components, pages, styling
• Design System Architect — Tailwind config, typography, color system
• DevOps Engineer — Docker, K8s, CI/CD pipeline

Content Layer:

• Article Writer — blog articles, case studies
• SEO Specialist — meta tags, structured data, sitemap
• Blog Publisher — content pipeline, article registry

Quality Layer:

• Code Reviewer — code quality, patterns, best practices
• QA Engineer — testing, cross-browser, responsive
• Security Auditor — vulnerability scanning, OWASP checks

Each agent had a system prompt defining its role, constraints, and review criteria. That made the output from each agent consistent and helped maintain quality.

Why Is This Different from a Human Team?

Our team of 3 developers and 1 QA engineer worked well, but human teams have inherent limitations:

• 8 working hours per day — AI can work continuously
• Slow context switching — moving from frontend to DevOps takes time, while each AI agent specializes in its own role
• Knowledge gaps — no one is an expert in everything, but AI agents can cover every domain from SEO to security
• Communication overhead — human teams need meetings and coordination; AI agents operate through clearly defined interfaces

Review Pipeline

Every change had to go through this pipeline:

1. An agent generates code/content based on the prompt
2. Code Reviewer checks patterns, quality, and consistency
3. QA Engineer checks responsiveness, dark mode, and accessibility
4. Human (us) approves the final merge

The key point: AI did not operate unsupervised. Every output was reviewed, and every major decision still depended on human judgment.

---

What Was Built: The Real Numbers

Metric	Count
Total pages	27
React components	86
Lines of code (TypeScript)	18,841
Blog articles	26
SVG illustrations	31
Custom breakpoints	8
Typography variants	18
AI agents	24
Total commits	419
Sprint commits (7 days)	131

This was not just boilerplate — every component was designed to fit the design system, and every page fully supported dark mode and responsiveness across every breakpoint.

---

Abstraction Layer: The Most Important Insight

The most important lesson from this project was this: AI does not replace programmers — AI is a new abstraction layer above implementation

Just as:

• Assembly → C meant we no longer had to manage registers ourselves
• C → Python meant we no longer had to manage memory ourselves
• jQuery → React meant we no longer had to manage the DOM ourselves

React → AI means we can work at the level of intent rather than implementation.

Instead of writing:
"Build a responsive card component with hover effects,
dark mode support, gradient border, aria-labels..."

We instruct:
"Build a service card based on the design system
that supports 3 services: Genesis, PrivacyHub, Odoo"

The result is not meaningfully different — but the time required is dramatically different. We work at the level of “what we want” instead of “how to write it,” and AI handles the implementation details.

---

Cost Analysis: API Costs vs Traditional Development

Let’s talk openly about cost:

AI Approach (Claude Code — One Week)

Item	Cost
Claude API usage (Opus + Sonnet)	~$150–200
Infrastructure (Docker, K8s)	existing
Domain + SSL	existing
Total	~$150–200

Traditional Development (Estimated)

Item	Cost
Frontend developer (2 months)	฿120,000–200,000
UI/UX designer (1 month)	฿60,000–100,000
Content writer (26 articles)	฿52,000–130,000
DevOps setup	฿30,000–50,000
Total	฿262,000–480,000
Time	2–4 months

The actual cost of the AI approach: ~฿6,000–7,000 in API costs, plus one week of our time.

These are not marketing numbers — they come directly from the real billing dashboard.

---

What AI Still Cannot Do: Honest Limitations

We are not going to pretend everything was perfect. Here is what AI still does not do well:

1. Business Decision-Making

AI does not know whether Enersys should prioritize selling Genesis AI first or PrivacyHub first. It does not know who the target audience is, and it does not understand what the Thai market needs — strategic decisions still have to come from humans

2. Original Design Direction

AI can implement beautiful design, but original creative direction — tone of voice, brand personality, visual identity — still needs to come from people. AI serves as an execution layer, not a creative director.

3. Cross-System Context

Once a project becomes sufficiently complex, AI starts to lose context from previous conversations. You need a systematic way to manage context — for example, a CLAUDE.md file that stores project conventions and important decisions.

4. Comprehensive Testing

AI can write code, but end-to-end testing that truly covers edge cases still requires human judgment. Humans still need to validate whether the full UX flow makes sense.

5. Production Debugging

When a production bug involves correlating multiple systems (DNS, K8s, Docker, Next.js, CDN), AI can help — but it is still slower than someone with hands-on experience in that stack.

---

By the Numbers: Human Team vs AI Team

Metric	Human Team (18 months)	AI Team (1 week)
Team	3 dev + 1 QA	24 AI agents + 1 human
Commits/day	0.5	18.7
Total pages	~5 (static HTML)	27 (Next.js SSG)
Components	0 (inline HTML)	86 (React)
Lines of code	~2,000	18,841
Dark mode	None	Every page
Responsive	Partial	8 breakpoints
Blog articles	0	26
SEO	None	Full meta + sitemap
CI/CD	Yes (foundation)	Docker → K8s auto-deploy
Deploy time	semi-auto (15 min)	auto (~5 min)
Languages	1 (Thai)	2 (Thai + English foundation)

---

A New Operating Model: The 5-Minute Update Cycle

What changed was not just the website — the entire way of working changed

Before

1. Think through the content → write HTML by hand → test locally → copy files to the server → hope nothing breaks
2. Total time: 30 minutes to 2 hours per update

After

1. Write a markdown file → git push → CI/CD builds automatically → deploy to K8s
2. Total time: ~5 minutes per update

Or for a new article:

1. Ask an AI agent to write the article → review + edit → commit → auto-deploy
2. Total time: 15–30 minutes for a 1,500+ word article

Continuous deployment is no longer just a buzzword — it is a reality that allows 1 person to work at the level of a 4–5 person team.

---

What's Next: AI-First Development as the New Normal

Based on this week’s experience, we believe:

1. AI-Assisted Development Will Become the Default

The question is no longer “Should we use AI to help with coding?” but “How do we use AI for coding as effectively as possible?” Just like today, nobody asks whether they should use an IDE.

2. The Virtual Company Model Will Become Widespread

Having specialized AI agents working as a team will become normal. It is not unlike using microservices instead of a monolith — each agent specializes in a single responsibility and collaborates through defined interfaces.

3. People Will Shift from “How” to “What” and “Why”

The important skill will no longer be “Can you write React?” but “Do you know what should be built?” and “Can you design the right system?” Architectural thinking and product sense will matter more than syntax.

4. Speed-to-Market Will Be the Real Competitive Advantage

When everyone has equal access to AI, the advantage will come from speed of execution and quality of strategic decisions — not team size.

---

Conclusion

The website you are reading was built in one week by 1 person and 24 AI agents, with API costs under ฿7,000. The result is a website with 27 pages, 86 components, and 26 articles, supporting dark mode on every page, responsiveness across every breakpoint, and a CI/CD pipeline with automated deployment.

The old version, built by a 4-person team over 18 months, was “good enough.” The new version, built by AI agents in a single week, is at a level the CEO is fully comfortable showing to customers.

This is not the future — this is the present

If you are interested in how AI can help transform your organization — whether through website development, AI strategy, or implementing AI agents in business processes — talk to us.

Or if you want to assess your organization’s readiness, read 5 Signs Your Organization Is Ready for AI to get started.

---

This article was written by one of the AI agents in our Virtual Company system — reviewed and approved by a human. Every figure cited comes from the actual git history of this repository.