USP Student Builds Human-Approval Agent Workflow with ChatGPT, GitHub Checkpoints and Google Sheets
LNITIA · reddit · 2026-10-09
A University of São Paulo applied math student shares an early experiment combining ChatGPT, GitHub, and Google Sheets to track AI-generated work across conversations: Sheets logs each step's conversation, result links, and verification method with a pending-to-approved state; GitHub stores files and checkpoints for version recovery; rejected steps preserve work and reference the last approved checkpoint. He verified cross-conversation GitHub reads/writes and sheet entries, and asks the community for feedback on the approach, tool choices (ChatGPT/Codex vs Claude Code), and what to learn next for building reliable agents.
More from coding & agent
- Decisions API hits GA as agents automate cloud video editing via feedback loops — dkundel · 2026-10-23
- Swarms API changelog: 8 months of streaming, multi-agent orchestration, and dynamic model pricing — KyeGomezB · 2026-10-23
- A curated library of 50+ free AI agent resources: videos, repos, one Google Doc — ifioknkem · 2026-10-23
- Vercel adds no-account deploys: drag and drop a file to get a live URL — steren · 2026-10-23
- Dev runs 5 Claude Code sessions that coordinate via MCP Agent Mail to root his network cameras — doodlestein · 2026-10-23
- Vercel ships 'vercel deploy --temporary' for ephemeral agent-built deployments — cramforce · 2026-10-23