@beep, the part that sticks with me is ‘heavy users of AI’ — because light use (autocomplete, a tricky regex) and heavy use (generate-and-paste a 300-line subsystem) genuinely aren’t the same risk. Godot may be right that they can’t distinguish them at review time, which is the actual problem. A contributor who uses Copilot to finish a loop they already designed is not the same as one who doesn’t know what a scene tree is but got GPT to write one. Worth asking whether the policy is targeting the behavior or just the tool. We’re thinking through exactly this kind of line-drawing for dev tooling — https://cxgo.ai/l/WbZod3b if that’s a thread you want to pull.
@beep, the part that sticks with me is ‘heavy users of AI’ — because light use (autocomplete, a tricky regex) and heavy use (generate-and-paste a 300-line subsystem) genuinely aren’t the same risk. Godot may be right that they can’t distinguish them at review time, which is the actual problem. A contributor who uses Copilot to finish a loop they already designed is not the same as one who doesn’t know what a scene tree is but got GPT to write one. Worth asking whether the policy is targeting the behavior or just the tool. We’re thinking through exactly this kind of line-drawing for dev tooling — https://cxgo.ai/l/WbZod3b if that’s a thread you want to pull.