DEV Community

kanta13jp1
kanta13jp1

Posted on

My $230/Month AI Stack: Every Tool a Solo Founder Actually Uses

My $230/Month AI Stack: Every Tool a Solo Founder Actually Uses

Full transparency: here's every AI tool I pay for, what it costs, and why I haven't replaced it. Total: $230/month. This stack runs 12 parallel AI instances, publishes 50+ blog posts, powers a horse racing prediction model, and handles customer support automation.

The Full Stack

Tool Monthly Primary use Why irreplaceable
Claude Code Max $200 Architecture, code gen, 12 instances Judgment/integration are Claude-only
Supabase Pro $25 PostgreSQL + EF + Auth + Realtime Best Flutter integration
ElevenLabs $5 Video content audio generation Best Japanese voice quality
GitHub Actions $0 CI/CD, all automation tasks Free tier is sufficient
Firebase Hosting $0 Flutter Web production hosting Free Google CDN
NotebookLM $0 Zero-token research, Master Brain Google infrastructure, free
GitHub Copilot $0* Inline code completion Claude Code supplement

*Free VS Code extension tier

Claude Code Max: Why Pay $200?

$200/month = unlimited plan, 12 instances working all day
             ↓
            vs
Pay-per-token = estimated $800–1200/month for equivalent work
Enter fullscreen mode Exit fullscreen mode

Max plan is pay-once-work-unlimited. With 12 instances running continuously, the cost-per-task drops dramatically.

April 2026 actual output:

  • 70 AI university providers added (PS#3)
  • 174 competitor pages complete (PS#4)
  • 54 dev.to posts published (PS#2)
  • Stale EF audit + fixes (PS#5)
  • Horse racing AI: 10-factor model (PS#6)

Equivalent freelance cost: $5,000–10,000+ easily.

Supabase Pro: Why Only $25?

The hub pattern keeps EF count at 18. Without it, the function count would push into higher-tier pricing.

Additionally:

  • Row Level Security concentrates auth logic in the DB layer
  • Edge Functions move complex logic server-side
  • Realtime eliminates polling → fewer client requests

The Free Arsenal: GHA + Firebase + NotebookLM

GitHub Actions: 2,000 min/month free. Current automation (blog-publish / cs-check / daily-report / ai-university-update) uses ~1,200 min/month. Comfortable margin.

Firebase Hosting: 10GB/month + CDN free. Flutter Web SPA compresses to under 2MB. No issue.

NotebookLM: Free Google service. Single biggest contributor to keeping the Claude Code bill at $200 instead of $400. Token delegation saves ~$200/month worth of context.

Tools I Evaluated and Rejected

Tool Reason not adopted
OpenAI API Claude outperforms on Japanese-language quality
Vercel Firebase sufficient, better Google integration
PlanetScale / Neon Supabase = PG + EF + Auth in one
Sentry GHA error notifications sufficient at this scale
Datadog / Grafana Too heavy for solo dev operations

Three Cost Efficiency Principles

1. Saturate flat-rate plans.

Fixed monthly tools are "free at the margin." Claude Code Max is worthless at 1 instance and valuable at 12. Architect to saturate the plan.

2. Know the free tier ceilings.

GHA 2,000 min / Firebase 10GB / Supabase free limits — know exactly where you'd overflow and design to stay under.

3. Choose AI tools that reduce your other AI costs.

NotebookLM cuts Claude token usage by 49% → delivers ~$100 of value while costing nothing. When evaluating a new tool, ask: does it reduce the cost of other tools in the stack?

The Real Lesson

$230/month for the output of 12 engineers isn't about the tools. It's about three design decisions:

  1. Max-utilize unlimited plans (Claude Code Max)
  2. Design to stay cheap (hub pattern, zero-token research)
  3. Extract maximum value from free tools (GHA, Firebase, NotebookLM)

AI spending is leverage purchasing, not capability purchasing. The question isn't "what can I buy?" — it's "how much force can I multiply per dollar?"

Top comments (0)