I asked four AI agent setups for the same thing: a bottom sheet and a stack of swipeable cards, like in a mobile app, as one HTML page that works with mouse and touch, plus a 15-second video of someone dragging, flicking and dismissing them. Their results are attached as gestures-1 to gestures-4: a screenshot and the video for each. I'm not telling you which model made which. Be the judge. Which one feels most like a real native app when you touch it, and why? Rank all four. Look at the motion: does the drag track the pointer, does a flick carry its speed, do things snap and settle naturally, can you catch something mid-animation, and does anything look broken. Some of them followed Emil Kowalski's gestures skill; say whether that made a real difference. Deliver one self-contained HTML page: a two-line verdict at the top, then the rankings and short notes. Keep it short enough to read in two minutes. Don't stop to ask me questions; make sensible assumptions and say what they are.