The demo in one GIF
The one prompt
The prompt asked for a playable chess game with legal moves, turns and a board that renders in the browser. Nine agent steps later the game was running — the shortest of the three demos, and the cheapest.
What the agent did
- Planned the board, the move rules and the turn logic.
- Wrote the game as a small set of files.
- Opened it in the browser and played a few moves.
- Verified legal-move handling and turn switching.
- Reported the finished game with the run summary.
The honest numbers
| Metric | Value |
|---|---|
| Model | deepseek-chat |
| Agent steps | 9 |
| Tokens | 24.2k |
| Cost | ≈ $0.0079 |
The three demos, three builds
Each showcase page is a separate build: this chess game, the Google Maps clone and the App Store clone. Three prompts, three plans, three traces — all recorded end to end.
Run your own
Send the same one-liner to your agent and compare the trace with the Deep Dive view. New here? Follow the macOS install or the Linux install, then set a token budget before you experiment.
Frequently asked questions
Why only 9 steps?
Is this the App Store demo?
Will the game always be legal-move correct?