I'm a high-load systems engineer, and I shipped a small frontend without writing any of its code myself. An AI agent generated everything while I directed it. The app lets someone send a link containing a question with YES and NO buttons; the NO button dodges the pointer. It has been live for about two months.
The surprising part was that the code was not the difficult part. Designing and communicating the animations was. I repeatedly tried to describe what I imagined, but the results rarely matched. The hardest example is a black-hole effect where the button gets pulled inward, the screen spins, objects disappear into the hole, and fragments shoot back out. Even experimenting with physics-based motion did not capture the feeling I wanted, and I'm still not satisfied with the result.
The project also started spreading on its own after a few posts, with people sharing it and traffic and payments arriving even when I was not actively working on it. That unexpected feedback loop has been the most rewarding part.
For clarity, I intended the experience as a teasing toy rather than a serious date invitation. Users can close it whenever they want, every invite has a report control, and I added a test that sends twenty NO interactions in a row and fails if any of them becomes YES.
For anyone using AI coding agents to build frontend interfaces, what techniques help you communicate motion and animation ideas more accurately?
3 Answers
Use a visual reference whenever possible. A screen recording, GIF, or similar interactive example gives the agent concrete timing, easing, and sequencing to work from. Tell it to start with that reference and explain only what should be different. Agents are generally much better at modifying an existing motion than inventing a complex effect from abstract words.
Plan before you build. I usually open a separate conversation and ask the agent to create an implementation plan based on a vivid description. I go back and forth until the plan reflects what I mean, then hand that plan to another conversation that handles the implementation. Separating planning from coding has worked much better for me than trying to explain everything while the agent is already building.
That is pretty close to my workflow too. Planning first and implementation second seems to help with frontend work just as much as it does with a conventional project.
The problem is that words are poor at expressing a motion that exists mostly as a feeling in your head. A chaotic effect like the black hole can turn into something that feels like a basic slideshow transition after hours of tweaking. Visual references should help, and your automated test for twenty NO interactions is a smart way to verify that the joke stays consistent without turning into an accidental YES.
Exactly. I eventually realized I was not describing mechanics anymore; I was trying to explain a sensation of movement that only existed in my imagination. And thanks—the interaction test was meant to make the intended behavior unambiguous.
That is a useful approach. Finding an exact reference may not always be possible, but even a partial match would probably be easier to communicate than describing the whole effect from scratch.