Last updated: September 13, 2026
Same folder of research. Same one-paragraph prompt. Two models, GPT-6 Astra on high and Fable 5.1. The job: build an 8-bit game where you sit in John D. Rockefeller's chair and make his decisions. Astra shipped the better-looking game in one pass. Fable shipped the one I wanted to keep playing. Here is why.
The idea came from David Senra's Founders podcast. He reads a founder's biography and walks you through the decisions. I wanted to play them.
Not read about the family loan, take it or refuse it. Not read about the partnership, join it or go alone. And when you choose differently from history, the game tells you what actually happened.
Two ways to play: founder or investor. The difference is chess. The founder sits at the board and plays every move. The investor walks in once every four hours, looks at the board, and walks out.
The rough version I built first
Before the comparison, I had a first version. Quick, bad prompt, built with Fable. It gave me a chapter screen that felt like a Netflix episode list: a summary of each stage, a "year four" counter, a cash-to-bankruptcy meter I misspelled on the sketch.
Some of that was worth keeping. The chapter summaries. The cash meter. What I did not want: made-up years. If the game says year four, it should be 1863, because that is when the real decision happened.
So I sketched the screen I wanted. 8-bit, retro, one chapter of Rockefeller to start, real dates, decisions pulled from the biographies. Then I put the sketch, the research, and the first build all in one folder.
The prompt, bad on purpose
I cannot write prompts. I say that in the video and I mean it. The instruction was long, messy, and I left it that way because that is how I actually work. Roughly:
In this folder there is a full research pack for the game I want to build. Look at the existing build, figure out what the problems are in your opinion, then build me your own version. 8-bit retro style. Open it in a web app for me.
Both models got the same folder. Astra ran on high. Fable ran in the same folder at the same time.
One difference, and I want it on the record because it matters for the comparison: I told Fable to dig deeper. Scrape the books, the articles, the interviews about Rockefeller, what other founders said about him, and add that to the game. Astra did not get that instruction. So the prompts were not identical. Fable had more to work with.
Playing the Astra version
Astra won the first look. The screen was clean, the retro style landed, and it played like a game I would have paid for on a handheld in 1995. First decision: your father offers you a loan. Take it or not.
For the record, the real answer is take it. The way Ron Chernow tells it in Titan, William Avery Rockefeller, "Devil Bill," lent his son money when the banks would not. Rockefeller borrowed. I know this because I am a nerd about it.
Second decision: Maurice Clark brings experience, do you go into the partnership or stay solo? History says partnership. I chose it, locked it in.
Then I chose against history on the next one to see what the game does. It gave me a short "you chose differently, here is what actually happened" panel. Correct, but thin. It told me I was wrong. It did not tell me why.
Playing the Fable version
Fable's version looked worse. Plainer screen, less polish. I said so in the video, and I still think Astra's interface is the one to build on.
Then the first decision loaded. Same family loan. Fable called the chapter "Borrowing from Devil Bill" and included the part Astra left out.
Rockefeller's own account says his father advanced the money as a gift. The same father is on record saying he cheated his boys every chance he got, to make them sharp. Two sources, one decision, and the tension between them is the actual lesson.
I made a wrong choice on purpose. Fable recorded my prediction first, then showed me the real outcome, then let me carry on down the changed timeline instead of resetting me. "You chose differently, we will give you a chance."
That is the game I described in the first thirty seconds of the session. Astra built the game I sketched. Fable built the game I meant.
The verdict, with the caveat
Interface: Astra, and it is not close. Content: Fable, and that is partly because I told Fable to go read more. Would Astra have matched it with the same instruction? Maybe. I did not test that, so I am not going to claim it.
What I will claim: the model that went and read the biographies produced a better game than the model that produced a better-looking game. For a build that is supposed to teach, that is the whole ballgame.
Next steps I am still building: chat with the founder inside the chapter, using everything published about him. Alternate timelines that keep running after a wrong choice. And more founders, one chapter each. If enough people want it, I will open-source the repo.
Whose company should I do next? Tell me under the video: Astra vs Fable: Who Built the Better Game?
Which model is better for building games, GPT-6 Astra or Fable 5.1?
In this one test, Astra produced the better interface and Fable produced the better content. The prompts were not identical, so treat it as one data point. Astra is the front-end pick. Fable is the pick when the build depends on research.
What did the game actually do?
You play one chapter of Rockefeller's life as founder or investor. Each screen is a real decision with the real year. You choose, the game records your prediction, then it shows what history did. Choose differently and the timeline branches.
Where did the research come from?
A folder I prepared before the build: notes from the biographies, David Senra's Founders episodes, and articles. Fable was told to go further and pull from books and interviews on its own. Astra worked from the folder only.
Can I play it?
Not yet. It is a demo on my machine. If there is enough interest I will put the repo on GitHub. Say so under the video and I will count.
Read the next one → You sold a $20 box. Where is the $20?
Get the receipt → The Y channel. The build, the cost, the receipt. Two a week.
← All posts