Less than five hours. One large-language-model co-pilot. Zero hand-drawn sprites. That’s all it took for SPACE SHOOTER DX, a mash-up of Space Invaders and Galaga complete with parallax starfields, multi-type UFOs, Stranger-Things-inspired synth loops and a laser that goes pew instead of meh. Roughly three-fifths of the 2.3k lines of TypeScript Claude 3.7 Sonnet’s idea. The remaining code was me wrestling explosions, hunting down free art and chasing down an audio fade that refused to behave.
(What does the “DX” stand for? Claude never told me. “Deluxe,” “Developer eXperience,” “Doesn’t eXplode.” Pick your favorite. I am pretty sure that it doesn’t mean “diagnostics” in this context.)
Why an ’80s space shooter has (a little) R&D relevance
My brother, who has coded for decades, watched the game, saw me working in Replit, we gave Windsurf (another agentic coding application) a whirl and asked Claude there to build a three-body gravitational simulation, complete with documentation files. Within minutes, he had orbiting points, zoom, pan and energy plots involving 30 gravitational bodies. The biggest hurdles we ran into involved incompatible versions of libraries. Without those, the entire process would have been closer to maybe 10 minutes to create a basic gravitational (n-body) simulator.
A gravitational simulator in minutes
The bottom line is that the space shooter took substantially more time with a more tenuous R&D connection than a basic gravitational simulation app. Claude breezed through Runge–Kutta integrators and SVG plotting because the math is well trodden. Here is a 3-D React based version that Claude hacked together in about 10 minutes. Maybe about 15 if you factor in a wave of minute UI improvements.
While these are anecdotes, a GitHub and Accenture’s controlled study showed Copilot users finishing enterprise tasks up to 55% faster than manual coding. What’s more, 62% of gaming studios already use AI tools, mimicking the growing adoption of AI tools across the software ecoystem. Unity’s 2024 Gaming Report confirms mainstream adoption for rapid prototyping.
In R&D contexts involving agentic AI, humans must stay in the loop, stressed Joe Mullen, director of data science at SciBite, in an article in Technology Networks Informatics. “Agentic AI has huge potential to augment how scientists work, but its deployment is an endeavor that demands expertise.”
If an AI can go from zero to stable 30-body orbits before you have a chance to finish drinking a cup of coffee, imagine what it can do to that data-cleaning script you’ve been putting off since Q4 of 2024.
