He Spent Millions of Tokens on a Loading Animation Only He Would See

Egor posted a loading animation on X. The use case was very specific: his own internal tool, with only him as the user. The cost was also very specific: two prompts, millions of tokens, and a few minutes of iteration. The original post is here.

The comments first debated value for money. BreezeOg replied to him: “Millions of tokens for a loader only you will see lol.” Egor's response was calm: every time he sees it, it gives him an ounce of joy, and that's enough. In another reply, someone said it was worth it, and he replied, “Absolutely.”

Janusz Aureliusz zoomed out to a longer timescale. He said humans used to make chairs by hand—beautiful, art. In the 1850s, machines mass-produced chairs; the first few seemed like magic, like an incredible achievement. Now no one cares; chairs are everywhere. Egor's animation is like those first machine-made chairs. Five years ago it would have amazed people; now it's just something machines can do, and in a few months no one will notice. Egor asked him in return: if an artist designs a beautiful chair and a factory manufactures it, is it not art just because a machine assembled it?

That rhetorical question did not end the discussion. Sof said the animation was too mesmerizing and might not be a good idea in a personal app—he would watch it all day. Ege noticed another layer: an unlimited-token plan should be nice. materkel left just one line: worth it.

tokenbender quote-tweeted it with a colder judgment. He said that creativity becoming the biggest differentiator—an expectation still hanging over this year—can now be completely crossed off. The quote-tweet image is below:

In a reply, he added that ownership of the result matters more than ever. He then listed a few things he found not bold enough: Millennium Prize problems were solved this year, the RL space was flooded 100x by data/environment vendors, and video/image generation was solved by downstream JavaScript. The tone was disappointment. The original post is here.

Putting these together, Egor's loader looks like a very small test case. Generation cost has become low enough to make an animation for an interface only he sees. Before, this required an animator, a frontend developer, and back-and-forth communication; now it takes two prompts and a few minutes. But whether it's worth it did not thereby become a pure compute question. BreezeOg was calculating exposure, Egor was calculating an ounce of joy, Janusz was calculating historical depreciation, and tokenbender was calculating who owns the result.

The machine-chair analogy is clever, but it misses one detail: the first machine-made chairs were also designed by someone, someone decided to produce them, and someone was responsible for the finished product. Today's loader is the same. The model can generate the animation, but the person who decides what to make, why to make it, and who it's for is still Egor. The cheaper generation gets, the more expensive that decision becomes.

In a few months, no one may notice animations like this. It was never meant to be noticed in the first place. Egor will see it, and that's enough.