DavidKPiano: Agents write better code than you — but you still must read it
DavidKPiano · x · 2026-09-15
David Kiano argues the real risk with AI coding agents isn't code quality — it's that agents make assumptions when information is missing and bake them into the code, producing bugs the agent doesn't even recognize as bugs.
- Even forcing the agent to ask 200 questions won't eliminate assumptions, since some things only become knowable at implementation time
- Code review agents review code at face value but rarely catch mismatches between the code and your intended behavior
The only fix: read the code yourself. Skim the tests, the core logic, or have the agent summarize it — just do a little human review.
More from coding & agent
- FlyWire publishes DIY guide to simulate a fly brain with ~160k neurons — patrickmineault · 2026-09-15
- 'Relay': a lightweight Codex skill for thread-driven multi-agent orchestration — daniel_mac8 · 2026-09-15
- Relay: open-source Codex skill hands bounded jobs to fresh tasks with one-message callbacks — daniel_mac8 · 2026-09-15
- Bolt Forge launches free until Oct 14 with GLM, DeepSeek and Kimi plus up to 50x more usage — HeyAmit_ · 2026-09-15
- A Practical Guide to Building an Agent Harness From Scratch — omarsar0 · 2026-09-15
- How to build an agent harness from scratch: LLM module, tools module, and the ReAct route — ryunuck · 2026-09-15