Capitalist Compounder Version 1.10 Game

For several years, in order to keep an eye on system capabilities and where I think the societal A.I. investments are going (especially as it may or may not represent a threat to the existence of moats for long-term investments), I had a series of tests – low-hanging fruit – I’d run to gauge the total cost, efficiency, and repeatability of output. The advances, as you know, have been staggering in recent quarters so we might as well be talking about ancient history at this point. By the time I write this on August 28th, 2026, these tests are basically child’s play to the more capable models.

Anyway, going back in time, one of the tests I developed for myself to gauge the increasing sophistication during that moment in time involved the creation of a simple 1970s or 1980s-style arcade game that I called “Capitalist Compounder”. It was really a sort of interactive case study. I acted as if I were a new owner of a video game company who had no coding experience or knowledge and was giving an assignment to an engineer to mock up a prototype for me so we could ship to arcades across the country in the pre-Apple II era. I was detailed about the exact structure, rules, and mechanics I wanted. Specifically, the game must:

• Be completely self-contained in an HTML file that could be run from any browser with no external references or asset dependencies;
• Have specific physics regarding placement of enemies, the bounce rate off enemies, and the speed and size of enemies, especially how those factors interacting with a sense of “fairness” so the player never doubted they were always capable of winning;
• Have specific mathematical compounding mechanics that mimicked real-world investing mechanics with different sub-types of contributions to the compounding base;
• Required virtually no load time;
• Had a repeatable musical theme that could be played endlessly and easily remembered;
• Accelerated the character to increase difficulty over time;
• Had a magical top-hat that represented an unrelenting bull market during which the player could crush competitors and the hyper-compounding rate temporarily accelerated;
• Was brutally unforgiving; a single mis-step and it’s game over; and
• Had a clear scoring system so that a person could compete against himself or herself, or others, knowing exactly where they stood.

In addition, one of the test conditions required that I would not alter, or correct, a single character of code by hand. Remember, the thesis was it was the pre-Apple II era, I somehow had found myself the owner of an arcade video game company, and I knew nothing about technology. All I could do was provide creative design and instructions on mechanics.

It was a lot of fun, and worth the credits burned, since I viewed it as an important insight into the landscape. The systems, as many of you know, are infuriating in that they are never entirely repeatable. They defy explicit instructions including prohibitions against altering specific features or lines of code. When they do fix something, it is more akin to spaghetti than prose or art making it a hot mess under the hood. Sometimes, the results were just downright comical. (In one derivation, the Capitalist in Capitalist Compounder kept flying off the screen like that famous kid off the cliff in Walker, Texas Ranger, but only on Android devices due to a glitch in which acceleration was set at the maximum rate the CPU could handle. Over and over again it would be fixed only to be re-introduced anew in the next line of revisions that were entirely non-related. It’s like the model was compelled to keep to making the Capitalist reach sonic-boom territory.)

That’s all besides the point. A funny thing happened that has me thinking a lot about game design. Capitalist Compounder 1.10 somehow became a hit in the Kennon-Green household. The boys are obsessed with it as I had them test it during different phases. Even now, as I wrote this post and they walked in to give me a hug before leaving for school, they saw it on the screen and begged to play for a few second before Aaron had to usher them into the car to not be late.

• It doesn’t matter that the graphics look 50 years old.
• It doesn’t matter that there is no complex lore or narrative.
• It doesn’t matter that it could run on the computing equivalent of a potato with some wires.

The basic structural design of the game captures something in its ability to produce dopamine hits. You play for a few seconds or a few minutes then walk away, no consequences. Then, you play again whenever you feel like it. Yet, in contrast, the video game industry has been overrun by nine-figure budgets, years of delay, player disappointment, and then, when something does work, inevitably the corporate parent disbands the team responsible. It’s enraging. It takes decades to come out with sequels to games like Fable or Skyrim. This is a human management problem, not a technological one. Games don’t have to be simple, but even the most complex somehow capture this magic; e.g., the last ones I played that seem to “get it” were Factorio and Baldur’s Gate 3. Throughout my life, SimCity 2000 was another illustration. There was something inherently human, and magical, about the game design. I suspect everyone is simply chasing the billions of dollars Genshin Impact has generated for its parent company.

This will continue to be one of my tests and, I suspect, in a few years, it will look like a finished Nintendo game as each iteration improves. Again, this version is already old and the newer models are capable of so much more. The point, though, is manifold. These systems are still hot messes. Prototyping are where they really shine. The design is the secret sauce, the code is merely the execution.

(Separately, as it pertains to enterprise software, one of my recent private client letters at Kennon-Green & Co. actually goes into a much larger discussion of how I see this interacting with the economy, investments, and job market. I think despite a lot of short-term upheaval, really talented senior engineers are going to have a field day cleaning up some of the absolute garbage code that is being injected into corporate systems right now by automated tools. The contingent liability building up is wild. I suspect more states will follow California’s model after it enacted AB 316, which removes liability protection from companies that deploy A.I. tools and then claims they (the company) aren’t responsible for its output or consequences. Essentially, it forces companies to treat software systems as if it were an employee so the costs of mistakes roll uphill to the actual executives, managers, and owners.)

Capitalist Compounder Version 1.10 – the game and its cover art – were created with a mix of tools that included ChatGPT’s 5.1 Thinking Model, Claude Code Sonnet 4.5, and Midjourney 7. Even this version, which again is now ancient due to how rapidly the tech has progressed, has some odd behaviors in it; e.g., depending upon the browser or hardware you are using, the enemy spawn rate seems very different from day to day. In a browser, you have to have the mouse inside the playing window itself otherwise spacebar (or mouse clicks) won’t work, which makes sense since the browser takes precedence and assumes you want to scroll the page when pressing spacebar by itself. I’m putting it up here as a reminder to myself of an interesting moment when technology was changing, sort of like the 2007-2008 era with smartphones. I want to look back and be able to remember it.

During the period in which I was working on these, I’d sometimes put a few of these tools in a segmented section of this blog, which almost no one noticed – https://www.joshuakennon.com/tools – that includes games, an amortization calculator, etc. Basically, it gave me a way to test a self-hosted version of the output isolated from the actual blog contents.