For those who just want to try the games, here they are.
The experience is nicer on a computer display than on a phone, just because the computer display is larger.
As a test of agentic programming, I decided to install the agent harnesses1 from OpenAI, Grok, and Gemini. Then I used all three harnesses, mostly OpenAI's Codex, to develop a circle pong game. It's just something I had seen written in Java (not JavaScript) before and thought it was kind of cool, and would make a good first project as it's engaging and easy to understand.
Agentic programming feels to me a lot like supervising a small group of software engineers. I started with a simple, open-ended request (prompt) to all three harnesses: Gemini, Grok, and Codex.
Write me a circular pong program that I can play by moving a curved paddle around the circumference of a circle to keep the ball in the circle.
Then I took some of the ideas from the games they created and added some of my own ideas: the slider panel, game physics for the paddle,2 specifications for the controls, and aesthetics. Then I executed at least 30 prompts, mostly to Codex, to get the games exactly the way I wanted them. Some were simple prompts and some more detailed. Here's an example of a prompt to make the game playable on a phone as well as a computer.
Update circle-pong.html to support touchscreen devices such as iPhone and iPad.
1. Make a single-finger touch-and-drag perform the same game-control behavior as the existing mouse drag.
2. Preserve all existing mouse and keyboard controls.
3. Make the existing Start Game/Try Again button work with a touchscreen tap as well as a mouse click.
4. Add a Pause button above the "Press [Space] to Pause" text that can be activated by touchscreen tap or mouse click. It should perform the same pause/resume function as the existing Space key.
5. Preserve the existing Space-key pause/resume control.
6. Do not change any other behavior or appearance except what is necessary to add these controls.
Here's the prompt to replicate the retro styling of one of the games created by Gemini. Once I got circle-pong.html exactly the way I wanted it, I copied it to circle-pong-retro.html and sought advice from Chatgpt chatbot on how to write the prompt to OpenAI's Codex harness.
Update circle-pong-retro.html so its visual appearance matches circle-pong-retro-gemini.html as closely as practical.
Use circle-pong-retro-gemini.html only as the visual reference. Transfer its retro styling, including colors, typography, panels, buttons, sliders, scanline effect, arena, paddle, ball, and other visual elements.
Preserve all existing functionality, game behavior, physics, controls, responsive/mobile behavior, text content, settings, and game rules in circle-pong-retro.html.
Do not replace circle-pong-retro.html with the Gemini implementation or copy its game logic. Modify only CSS, presentation markup where necessary for styling, and canvas drawing code required to reproduce the retro appearance.
The results almost always came out exactly as I'd intended.3 Just a few bugs that were fixed on the first try. It's amazing what these harnesses and agents can do! Each file is about 700 lines of code. They are written in HTML, CSS, and JavaScript. It would take me a couple of years to learn enough JavaScript and game architecture to write these, if I could do it at all. But I could do it in a day with coding agents.
1. The harness is software that encompasses and manages the AI agent. You give the harness a task in plain English, and it manages the agent as it plans the work, executes the plan, tests the results, and tries again when necessary. Technically the human interacts with the harness and the harness interacts with the agent.
2. In real-world paddle physics, if the ball always initiates from the center of the circle, then when it hits the paddle, assuming there's no spin or paddle friction built into the game, the ball will bounce straight back to the center of the circle because the paddle is an arc along the circle. That's pretty boring. Gemini wrote its pong game that way. Grok used what's know as game physics. If the ball strikes the paddle some distance from its center, call that distance off-center, then the ball deflects at an angle away from the center of the circle, proportional to the off-center distance. If the ball strikes to the left of center, it will deflect left. If the ball strikes to right of center, it will deflect right. Grok used such strong game physics on the paddle that if the ball struck near the same edge of the paddle a couple time in a row, it would skip round the boundary of the circle. The best one could do is chase the ball with the paddle until it eventually escaped. Codex used game physics that was more moderate, so I asked Codex to use that one. A future update might be to increase the strength of the game physics a little more.
3. I'm not counting the time I asked Codex, all in one prompt, to take Grok's initial pong game circle-pong-grok.html and port over all of the controls and features I had been developing in circle-pong.html, because I liked the aesthetics in circle-pong-grok.html. That's of course a terrible idea, but I wanted to see if Codex could do it. It replicated the slider panel with the PLAY AGAIN button but placed it only half way into the circle. Even if I wanted to follow this path, it's better to request one feature at a time so that any bugs are easier to track down. I bailed on this path, and instead asked Codex, one feature at a time, to add a few aesthetics to circle-pong.html that were presently in circle-pong-grok.html: appearance of circle perimeter, appearance of paddle, comet tail on the ball, and a few other niceties.