Decide how far to delegate on an in-house framework

Your team built its own stream-processing framework years ago. It has unusual rules for time windows, late events and retries, and it exists nowhere in public code. A new hire proposes building a major feature on it with an AI agent. Their plan is the one they used on React apps: describe the feature and let the agent write most of it.

Is this a good fit for heavy delegation? Decide how you would use the agent here, and what changes because it has never seen this framework.