Non-coder runs hundreds of thousands of AI-written lines on a phone via zero-trust Termux workflow
amadale · reddit · 2026-09-03
A self-described non-programmer Korean author explains his "Discussion Coding" methodology: he maintains hundreds of thousands of AI-assisted lines purely on a phone inside Android Termux, built not on trust but on Zero Trust.
Key points:
- Human as router and gatekeeper: he manually copy-pastes between multiple cheap chatbots from different countries, compares their reasoning himself, and never lets AI execute code directly — approval goes through him.
- Deterministic verification: since he can't review code and won't trust AI self-checks, everything must pass through his hand-built "Garlic" system in Termux, where machine-produced RAW data is the sole standard for the next action.
- Physical isolation: all execution happens in the Termux sandbox.
- Traceable state: integrity SHA hash chains, checkpoints, and event sourcing transition state step by step.
- No tool? He builds one immediately; failures become homework until solved.
He admits his limitation ("a blind man with open eyes" on code) but claims the distrust-driven pipeline has kept the system nearly error-free.
More from coding & agent
- Claude's virtual machine takes up a whopping 22GB of disk space — DevDminGod · 2026-09-03
- ContextPilot uses RL to teach agents proactive context management, 69.4 vs 45.9 — rohanpaul_ai · 2026-09-03
- Dev Predicts Context Compaction Will Be Replaced by Agentic Session-History Retrieval — mayfer · 2026-09-03
- Multi-agent debugging pain: per-agent logs can't show which output changed the next agent's decision — mageblex · 2026-09-03
- PR adds ready-made JS and Python computer-use environments to OpenAI's CUA sample app — ChrisGPT · 2026-09-03
- 2-hour vibe-coded local agentic browser clones Opera Neon's $19/month features with MCP — comperr · 2026-09-03