A live session where Shahar built a 2D game using Unity AI Assistant, Claude Code, and Rider — testing which tool wins.
An informal live experiment — Shahar explored Unity AI Assistant in real time with viewers, no fixed plan
Setup required opening Unity 6, Rider, Warp terminal, and Gemini in parallel; Unity AI signup asked for billing details
Game concept chosen with viewers: a 2D magnet/planet puzzle game (hyper-casual) using Kenney assets
Dual-tool strategy: Claude (via Rider/Claude Code) for architecture and code, Unity AI Assistant for scene/hierarchy
Unity AI Assistant was very slow and required frequent manual confirmations for every action (file saves, screen captures)
Unity AI wrote low-quality code: OnTriggerEnter with CompareTag, no namespaces, poor performance patterns
Claude produced better code and managed git; MCP for Unity was installed via Claude Code — one line in package.json
Main conclusion: Unity AI Assistant is disappointing — useful only for minor tasks, not autonomous scene building
Better stack for Unity AI dev: Claude Code + Rider + Warp; Gemini and Codex also show strong code architecture
End result: a basic, partially working game (magnetic ball, planets, levels) — unpolished. Blog post planned.
Opening — informal experiment intro, blog post planned
Viewer poll: 2D or 3D? → 2D chosen
First opening of Unity AI Assistant
Gemini brainstorm: 10 hyper-casual 2D puzzle ideas
Unity AI Assistant package install + service signup
Dual-tool strategy explained: Claude + Unity AI
Unity opens; Rider set as script editor
Warp terminal intro — managing multiple agents
Project pushed to git, Unity AI connected
Unity AI starts coding — constant approval prompts begin
First frustration expressed: too slow, poor UX
Key question: why use Unity AI if Claude does it better?
MCP for Unity demo — one-line install via Claude Code
Unity AI installs MCP but scene won't run
Switching to domain-based commits; CLAUDE.md initialized
Play mode enabled manually — game partially works
Level layout looks reasonable; object scales corrected
Key insight: Claude + custom MCP would have been better
Final verdict: Unity AI writes code but can't test/run scenes autonomously
Unity AI succeeds with very precise instructions — but that's not practical
Session end — disappointment expressed, blog post planned
Vibe Coding 101 teaches you to build Unity games with AI — from zero to a finished product, using the Director vs. Typist method.
Learn About the Course →