How to Tell Whether an AI Game Bug Comes From Code, Scene State, or Engine Setup
Debug by layer. Reproduce the symptom, inspect runtime state, then classify whether the logic is wrong, the scene/data is wired incorrectly, or the engine/project configuration is causing the behavior. Do not let the AI edit code before the layer is identified.
Debug by layer. Reproduce the symptom, inspect runtime state, then classify whether the logic is wrong, the scene/data is wired incorrectly, or the engine/project configuration is causing the behavior. Do not let the AI edit code before the layer is identified.
Code is only one source of game bugs
A correct script can fail because a serialized field is empty, an object uses the wrong layer, a signal/event is not connected, an input action is missing, or a scene/prefab has stale state. AI agents often over-focus on code because code is the easiest artifact to inspect.
Use a layer triage
- Code layer: pure function/state transition is wrong under known inputs.
- Scene/data layer: script is correct but references, values, hierarchy, or connections are wrong.
- Engine/project layer: input map, physics settings, build target, permissions, import settings, or platform behavior differs.
Ask for evidence that discriminates between layers
Log the value entering the suspect code, inspect the bound object/reference, and compare project settings or platform behavior. A good diagnostic step makes one layer more likely and another less likely; random edits do not.
Patch the layer that failed
If a node path is wrong, changing movement math is noise. If the project input map is missing an action, rewriting the controller creates debt. Keep the repair aligned with the evidenced layer, then rerun the original reproduction.
Working template
SYMPTOM: [...] REPRO: [...] CHECK 1 CODE INPUT/OUTPUT: [...] CHECK 2 SCENE/DATA WIRING: [...] CHECK 3 ENGINE/PROJECT SETTINGS: [...] CLASSIFY LAYER BEFORE EDITING. PATCH: smallest change in the failed layer; rerun same repro.
Source and verification note
This guide was checked against Microsoft Research: Vibe coding and Microsoft Research: Good Vibrations? and Vibe Coding in Software Development: Multivocal Review and updated Sep 5, 2026. The research describes patterns and risks rather than guaranteeing outcomes for every project; validate the workflow in your own codebase.