Workflows are JSON — structured data. n8n is the on-ramp; typed Python is the destination. This is the receipts trail behind the automation thesis.
The thesis: Workflows are JSON — structured data. n8n is the on-ramp; typed Python is the destination.
Receipts: 13 production flows · 2,053-workflow corpus studied · 20+ APIs orchestrated · self-hosted (Docker/Postgres/Caddy/DO) · converted to typed Python · taught in a workshop.
| Flow | Nodes | Integrations |
|---|---|---|
| AI Clone Automation | 59 | Airtable, Blotato, Code, Elevenlabs, HTTP, IF, OpenAI, Openai |
| AI Social Video Generator | 39 | Blotato, Code, Google Sheets, HTTP, Kling, OpenAI, Openai, Set |
| Blotato 1 Post Everywhere | 21 | Blotato, Google Sheets, Schedule, Set, blotato, stickyNote |
| HeyGen Clone | 20 | HTTP, IF, agent, lmChatOpenRouter, manualTrigger, stickyNote, wait |
| (Temerarii) AI Avatar for Trending News | 27 | Blotato, HTTP, Heygen, OpenAI, Openai, Schedule, Set, agent |
| UGC with nanobanana | 74 | Code, HTTP, Merge, Openai, Set, Switch, aggregate, chainLlm |
| Automate video creation with Veo3 and auto-post to Instagram, TikTok via Blotato - vide | 29 | Blotato, Google Sheets, HTTP, Openai, Schedule, Set, Veo, agent |
| Veo3 in n8n | 11 | Google Sheets, HTTP, Openai, Schedule, Veo, agent, lmChatOpenAi, toolThink |
| Viral news to AI avatar | 35 | Blotato, HTTP, Heygen, IF, Merge, OpenAI, Openai, Perplexity |
The n8n JSON → Pydantic-AI conversion (typed · testable · git-native · agent/MCP-native): .gitkeep, viral-news-avatar. See workflow.py · models.py · config.py · skills.md.
| Option | Pros | Cons |
|---|---|---|
| n8n | fast to learn · visual · huge node library · API-able · great teacher | JSON sprawl · hard to test/version/diff · runtime lock-in · brittle at scale |
| Pydantic Python | typed+validated · unit-testable · git-native · portable · agent-native · scales | needs engineering · no visual canvas · higher upfront build |