Eng lead: AI coding in team repos requires human gatekeeping for all serious architecture
sean_from_earth · x · 2026-10-15
An engineering lead shares real-world lessons from running a multi-contributor codebase where each developer uses AI agents: every piece of serious architecture must go through him or his best staff engineer, no exceptions. He admits this makes him work harder than pre-AI coding, but every time the rule was relaxed, AI slop crept in. He blames himself for insufficient team training and missing systems, yet stays too busy to fix them. The thread context notes a gap in current discourse: people either do solo vibe coding or AI labs Ralph-Wiggum their way through giant codebases—almost nobody discusses multi-human, multi-agent collaboration in non-commodified domains.
More from coding & agent
- Eazo thread details: full-stack workflow, 6+ design directions per prompt — HeyAmit_ · 2026-10-15
- Testing Eazo: plain-English prompt to a full app with zero code — HeyAmit_ · 2026-10-15
- Builder feeds a deliberately vague brief to Eazo AI agent, gets a full app — HeyAmit_ · 2026-10-15
- Composio Instant Lets AI Agents Pay for Their Own Tools, With $1M Giveaway — PolarBearby · 2026-10-15
- Blocked by anti-bot sites? Plug a $42 KVM into an old PC and let Grok fly past detection — Scobleizer · 2026-10-15
- Free open-source Claude Code course: 15 modules, 9+ hours, zero cost — ZabihullahAtal · 2026-10-15