The agents built the rooms.
A person connected the clues.

Memory Dive began with its developer’s own book. Turning it into an investigation was a different problem from illustrating its scenes: players could explore in different orders, and a character should not know a clue before the player discovers it.

In our interview, the developer describes AI agents working across code, Blender-based 3D and visuals, while he directed the production and corrected inconsistencies. Asked about those production tools, he put one part of the workflow plainly:

I didn't use a code editor.

TheAuthorBTLG_ · on the tools used for Memory Dive

This complete sentence is reproduced verbatim from the approved interview; the rest of the conversation is not reproduced here. It does not mean that no code was written or that human direction disappeared.

The game behind the workflow

The important player-facing idea is an unreliable reconstructed memory. You inspect objects and connect evidence, so a mistake in what a character knows could undermine the mystery. The developer’s account makes the distinction clear: generating material and connecting it into a coherent investigation were different tasks.

The Steam disclosure supports extensive AI assistance across code, text, 3D, artwork and audio, under human review. It explicitly excludes generation during gameplay and external AI-service connections.

Read the full published interview ↗

The original published interview was edited and approved by the guest. The single quotation on this page also matches his original wording.

Explore Memory Dive and its AI production disclosure →