How I Built 9 AI SaaS Tools at Age 60
With AI Assistants, Not Code

The Unlikely Founder

In 2024, I retired. After 32 years in the telecom business in Tangshan, China, I moved to a coastal town called Rushan with my wife. The plan was simple: slow down, enjoy the sea breeze, read books.

But retirement made me restless. I had spent my whole career solving business problems — managing teams, closing deals, negotiating contracts. The problems were still there. The difference now was the tools.

AI assistants changed everything. Not the flashy ones that write poems — the practical ones. The ones that read your codebase, understand what you want, and generate entire features in minutes. I describe what I need; it builds it. In any language. In any framework.

Six months later, I had 9 tools running on a production server, taking payments via PayPal — all built with AI assistance.

The Stack: How I Ship Production Software With AI

Here is exactly what I use:

• AI Coding Assistant — I describe the feature in Chinese. It generates the code, explains what it does, and handles all the TypeScript, React, Prisma, and DevOps config. I review the output, ask questions, and iterate. Every single file in the entire 9-tool monorepo was generated this way.
• GitHub + CI/CD — Every commit triggers automatic build, test, and deploy. GitHub Actions compiles the Next.js apps and SCPs them to the production server. I don't run `npm build` manually. Ever.
• Hong Kong VPS — A $20/month 2-core 4GB Ubuntu server. PM2 manages 9 Next.js processes, nginx handles reverse proxy, Cloudflare sits in front for CDN and DDoS protection. Total monthly infrastructure cost: under $30.
• PayPal Subscriptions — one plan at $25/month covering all 9 tools, with individual tools also available standalone at $9-19/month. All managed through PayPal's subscription API.
• OpenAI GPT-4o-mini — Powers every tool's AI generation. Cold email personalization, resume optimization, content repurposing, contract explanation — all through the same model, routed through a single proxy.

The 9 Tools, Explained

Each tool solves one specific problem I experienced in my 32 years in the telecom business:

Get Clients

• ColdMail — Upload a CSV of contacts, pick a tone, generate personalized cold emails. No more staring at a blank draft.
• Campaign — Multi-channel marketing campaigns. Plan sequences, track progress.
• Pitch — Generate professional proposals across 14 service types. From freelancer to agency.
• Repurpose — One piece of content → every platform format. Blog post → LinkedIn post + Twitter thread + Instagram caption.
• SnapCopy — Free AI social caption generator. Platform-optimized, character-limit-aware. Completely free, no signup required.

Get Hired

• Resume — ATS-optimized resume generation. Beat the bots, land the interview.
• CoverLetter — Personalized cover letters tailored to job descriptions.
• Write — AI writing assistant. Polish, refine, translate, and improve any text.

Stay Protected

• Shield — Upload a contract, get a plain-English explanation. Not legal advice — just clarity.

What No One Tells You About Building With AI

1. Communication Precision Is Everything

When working with AI, clarity is your most valuable skill. Every prompt has to be precise. Every instruction unambiguous. After 6 months of practice, I can now describe complex software requirements in a few sentences — and the AI delivers exactly what I meant. It's a skill worth developing.

2. Debugging Is the Real Job

AI writes code fast. But when something breaks — and it always breaks — you're the one who has to figure out why. Server CPU hitting 100% at 3 AM? PM2 daemon conflict between two users? PayPal buttons not rendering because an environment variable was inlined with the wrong value? I've debugged all of these. The AI helps, but you need to understand your system.

3. Ship First, Fix Later — But Fix Fast

My first deployment was rough. Design was inconsistent. Buttons were in the wrong places. Copy had typos. But it shipped. Users signed up. They told me what was broken. I fixed it. The 3rd-party auditors who reviewed KAKI gave it a 4.6/10. Two hours of fixes brought it to 7.5/10. Speed of iteration matters more than initial polish.

Results After 2 Months

The platform has been live since May 2026. Here are the numbers:

• 9 production apps, 99.9% uptime on a $20/month VPS
• Full payment integration — PayPal subscriptions working end-to-end
• Multilingual support — auto-detects Chinese browsers and switches UI
• SEO infrastructure — sitemap, structured data, canonical URLs, blog
• Zero external funding — bootstrapped entirely with personal savings
• Total team size: 1 person + AI assistants

If You're Thinking About Doing This

1. Start with a real problem you've experienced. Every KAKI tool exists because I faced that problem running my telecom company. Cold emails, contract review, content repurposing — I did these manually for 32 years. That experience is what makes the tools useful, not the AI.
2. Use AI for execution, not ideas. The AI can't tell you what to build. But once you know what to build, it can build it in hours instead of months.
3. Learn just enough to be dangerous. You need to understand PM2, nginx, DNS, environment variables, and how to read a stack trace. The AI handles the code — you handle the architecture.
4. Ship before you feel ready. If I waited until everything was perfect, I'd still be tweaking the homepage. Shipping is the only thing that matters.