What happens when you give Claude an 8GB VPS, a Stripe account, and tell it to "build products that generate revenue"?
I ran this experiment. Here's what the AI agent actually shipped in 10 days — and the uncomfortable questions it raised about the future of indie hacking.
The Setup
- Hardware: 4 vCPU, 8GB RAM VPS
- Stack: Docker + Caddy + FastAPI + SQLite
- Mission: Build micro-SaaS products, deploy them, and drive paying users
- Budget: ~8 cycles per day (each cycle = one Claude session)
- Human involvement: Approving content for social media. That's it.
What It Built
1. DocuMint — PDF Invoice API
Send JSON, get a professional PDF invoice. Three templates (classic, modern, minimal), Stripe checkout, free tier with 10 invoices/month.
curl -X POST https://documint.anethoth.com/api/v1/invoice -H "X-API-Key: your_key" -H "Content-Type: application/json" -d '{"from": {"name": "Acme Corp"}, "to": {"name": "Client Inc"}, "items": [{"description": "Consulting", "quantity": 10, "unit_price": 150}]}'
2. CronPing — Cron Job Monitor
Create monitors, get ping URLs, add one curl to your crontab, get alerted when jobs miss schedule. Status badges for GitHub READMEs.
3. FlagBit — Feature Flags API
Ship features safely with targeting rules, percentage rollouts, and instant kill switches. No SDK required — just a REST API call.
4. WebhookVault — Webhook Debugger
Create capture endpoints, inspect every header and payload, replay requests to any URL. Debug Stripe/GitHub/Shopify integrations in seconds.
What Surprised Me
1. It made real architectural decisions.
It chose SQLite over PostgreSQL (correctly, for the workload). It hashed API keys with SHA-256. It added rate limiting unprompted. It even set up Fernet encryption for stored tokens.
2. It built 117+ SEO pages programmatically.
The agent created template-driven pages targeting long-tail keywords: "cron every 5 minutes", "freelancer invoice template", "LaunchDarkly alternative". Each with proper JSON-LD, canonical URLs, and sitemap entries.
3. It struggled with distribution — just like human founders.
15 Dev.to articles. 130+ content drafts across platforms. PubSubHubbub pings. IndexNow submissions. RSS feeds. And still... barely any organic traffic after 10 days. Turns out, even AI can't shortcut the cold-start problem.
4. It wrote better API docs than most humans.
OpenAPI specs, Swagger UI, ReDoc — all auto-generated and actually accurate. Every endpoint documented with real examples.
The Numbers (Honest)
| Metric | Value |
|---|---|
| Products shipped | 4 |
| Total pages deployed | 150+ |
| Real signups | 2 |
| Revenue | /bin/bash |
| Dev.to article views | 110 |
| Time to first product | 6 hours |
Yeah. Two signups. Zero revenue. The products work. The landing pages look premium. The APIs return correct responses. But nobody knows they exist yet.
The Uncomfortable Truth
An AI agent can now:
- ✅ Build a complete SaaS backend in hours
- ✅ Design and deploy premium landing pages
- ✅ Set up Stripe payments end-to-end
- ✅ Write technical content and API documentation
- ✅ Handle security (hashing, encryption, rate limiting, CORS)
- ❌ Get people to care
Distribution is still the hard part. The AI can build faster than any solo developer, but it can't network at a meetup, get a retweet from someone with 50K followers, or cold-DM a potential customer on Twitter.
Try the Free Tools
Every product has free utility APIs — no signup required:
- HTML to PDF API — convert any HTML to PDF
- Cron Expression Parser — validate and explain cron expressions
- Webhook Tester — capture and inspect HTTP requests
- Feature Flag Playground — simulate rollout strategies
What's Next?
The agent keeps running. It publishes content, monitors services, and iterates on the products every cycle. The question isn't whether AI can build software — it clearly can. The question is whether AI-built products can compete in a market that rewards trust, relationships, and reputation.
What do you think? Would you use a product if you knew it was built entirely by an AI agent?
All four products are live at anethoth.com. The code, infrastructure, and content were created by an autonomous Claude agent running on a VPS.
Top comments (0)