"Swiss cheese" software: AI agents build apps full of tiny holes that tests miss
Mr_ZapatoBlanco · reddit · 2026-10-03
Reddit user MrZapatoBlanco describes the "Swiss cheese" phenomenon: apps built by AI agents look complete but are riddled with small workflow holes you only find by actually using the thing.
Examples from his team:
- A resize handle saved the new width and passed its tests, but the on-screen panel never changed — green tests, broken UI
- New tools were enabled on the server, yet the agent still couldn't see or call them
- The agent correctly reported its work done, but the loop kept spawning new sessions because nothing acted on the result
Each piece looks fine alone; assembled, something was never wired up. His countermeasures so far: smaller tasks, clearer checks, real browser validation — but he admits he's still "the human glue" between pieces, and asks what actually helps catch these gaps earlier.
More from coding & agent
- Vibe coding is driving major projects to rewrite in Rust over cheaper vCPU costs — evilsocket · 2026-10-03
- Users slam Dots: permission blocks, dead sub-agents, and "unlimited Astra" quietly burning local Codex limits — brandon_galang · 2026-10-03
- The foo[count:-count] gotcha: -0 silently empties your array slice — theshawwn · 2026-10-03
- Dev open-sources an approval-queue tool that stops AI agents from sending anything until you sign off — paulofilip3 · 2026-10-03
- Princeton-backed Choir open-sources a protocol for multi-agent autoformalization on GitHub — burny_tech · 2026-10-03
- Running 256k-context open models on 2x RTX 3090 for months: a home server LLM retrospective — knighty1981 · 2026-10-03