Vibe coder can't read AI-generated code: black-box testing alone hits a wall
Ryan_yii · reddit · 2026-09-20
A vibe coding practitioner describes the core anxiety of relying fully on AI agents: he can only verify features via black-box testing and can't spot implementation details or hidden risks in the codebase.
He asks three questions:
- What pitfalls come from letting AI write code you can't read?
- Do experienced developers actually review every block AI generates?
- Is there a practical code review workflow for catching hidden bugs, performance bottlenecks, and security vulnerabilities?
His current flow — state requirements, let AI write, run and check — breaks down once projects grow to multiple modules. No answers in the post, but it crisply captures the central tension of vibe coding at engineering scale.
More from coding & agent
- Local Qwen 27B fixes commercial app bug on first try at ~2x Claude Opus speed — julianharris · 2026-09-20
- MIT Media Lab to Host ScienceClaw Hackathon on Building an Internet of Agents for Science — ProfBuehlerMIT · 2026-09-20
- Reddit debate: multi-agent systems still lack a production-grade communication substrate — One_Two_2229 · 2026-09-20
- David Khourshid: Model agent logic as state machines — it works extremely well — DavidKPiano · 2026-09-20
- AI agents hit the monitoring first: observability is inside the blast radius — victor_explore · 2026-09-20
- Production AI pipelines have ~7 stages, and only one is the model — Aiden_Tech_Ai · 2026-09-20