What it is
A latency demo comparing Jev, a typed-decision model, against chat LLMs on Vercel AI Gateway. Each model plays the right paddle in Pong; one decision moves the ball one segment. Built by Ably's co-founder.
How it uses Jev
The agent asks Jev for a move once per ball step, sending a ~125-byte JSON state of court, ball, paddle and interceptY. Jev returns a typed choice (up, down, stay) via experimental_evaluate, and the agent applies it to the paddle.
Primitives:choice
Technique worth stealing
Same serve, rules, question and Gateway key across lanes, so a lane's speed is purely that model's decision latency.
Try it
Play at jev-pong.ably.dev with arrow keys or W/S, or drag on a phone.
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricno0.33
- Shows a System One patternyes0.84
- Handles uncertaintyno0.04
- Measuredno0.14
- Runnableno0.20
- Worth recommendingno0.28
- Model replicano0.04
- Problem scopescore on a 0–2 scale0.01
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.