When AI enters a classic, what should it be allowed to change?

While building Odyssey, I kept returning to one design question: when AI enters a classic, what should it be allowed to change?

I chose a bounded system. The route across fourteen shores is fixed. A deterministic engine controls progress, choices, and endings. GPT-5.6 voices Homer and a small set of divine encounters, but each voice has limited authority. Penelope’s final recognition is authored rather than generated.

That makes the experience less open than a generic chatbot, but more faithful to the original work—and, I believe, more trustworthy.

I’m curious how other makers draw this line. In literary or educational products, should AI maximize freedom, or work within a clearly authored structure?

Where does personalization stop deepening the experience and start erasing the source?