<aside> 💡

tl;dr I asked 5 AI coding agents to develop a browser-based version of Euchre (popular midwestern card game) with a single prompt and their default models (no examples, edits, or modifications).

Only two of them created a playable version

</aside>

I. Introduction

I used a single prompt to create a playable game of Euchre through coding agents. No second prompts, no examples, and no edits. Zero-shot using the default model. This was an experiment to see how far the coding agents have come given the recent advancements in models.

<aside> 💬

“Create a new euchre game that can be played on a browser with 3 other CPU players.”

</aside>

II. Coding Agents

Gemini CLI

The Good ✅

1. Decent UI

The Bad ❌

1. The game wasn’t playable
2. All cards were face-up
3. No scorekeeping

v0 (by Vercel)

The Good ✅

1. Decent UI
2. Cards were face-down

The Bad ❌

1. The game wasn’t playable
2. Gameplay rules not enforced
3. No scorekeeping

Lovable

The Good ✅

1. The game was playable! 🎖️
2. All cards were face-down
3. Scorekeeping

The Bad ❌

1. Gameplay could be improved
2. Gameplay rules not enforced