I built a little game with GPT-6 over the weekend and called it Thalassa. When I shared it, about half the graphics was still missing. My plan was to return to it the following weekend and finish more of the presentation.
That is a good stage for a prototype to be visible. There is enough of a thing to react to, while the incomplete parts make it obvious that the work is still open. The important question becomes what the next version needs, rather than whether the first version looks ready for a store page.
A rapid game prototype has two kinds of missing information. Some is visibly absent, such as unfinished graphics. Some only appears through use: an action that is hard to understand, a transition that gives no feedback or a rule that becomes confusing after the first attempt.
I would separate those categories before asking an agent to "finish the game." A replacement image may improve presentation without making an unclear interaction easier to understand. Conversely, a small timing or feedback change can make a rough scene much more readable.
The next task should be small enough to judge. Complete one coherent passage through the prototype. Make the result of an important action visible. Resolve one place where the player cannot tell what happened. Those are better instructions than asking for a general layer of polish over everything at once.
Working quickly with a model also creates a selection problem. It can produce alternatives faster than I can meaningfully evaluate them. Adding every plausible suggestion would make the game larger without necessarily giving it a clearer identity.
I want the prototype to answer a few basic questions first. What does the player do repeatedly? What makes that action interesting? What changes as the player becomes more familiar with it? The graphics can then support those answers rather than become a separate collection of attractive assets.
Sharing Thalassa in this unfinished state was a way to mark a concrete piece of progress. The weekend produced something with a name and a place to open it. It did not settle all the design questions, and the missing graphics were still real work.
That is enough for a useful prototype. It gives the next weekend a starting point. Instead of beginning with a vague game idea again, I can return to a specific object and choose the smallest change that makes it more complete.