Jan 16
Loading video player...
Timestamps: 00:00 - Intro 00:50 - Test Introduction 02:09 - Prompt Overview 03:22 - Coding Model Settings 03:52 - Codex Product Idea 04:39 - Claude Product Idea 05:11 - Vibe Building Begins 06:16 - Codex 3D Product Design 07:01 - Claude 3D Product Design 08:05 - 3D Enclosure Results 09:35 - Agent CAD Designs 14:46 - Physical Building Steps 15:15 - Codex Software Test 16:58 - Building Codex Device 19:32 - Building Claude Device 23:13 - Claude Code Result Overview 25:38 - Codex Result Overview 26:38 - Claude Code Product Detail Look 28:04 - Codex Product Detail Look 29:04 - Having Them Check Their Work 30:34 - Experiment Summary 31:16 - Closing Thoughts AI Integration & Consulting https://bijanbowen.com Join The Discord: https://discord.gg/hfaR2exy7S In this video, we put GPT-5.2 Codex and Claude Code Opus 4.5 into a head-to-head experiment focused on vibe coding a real physical product. Rather than limiting the comparison to abstract prompts or benchmarks, we task each model with ideation, 3D design, enclosure planning, software development, and real-world build considerations. We begin with shared prompts and configuration settings, then let each model independently propose a product concept. From there, we move into 3D product design, CAD-style enclosure generation, agent-driven refinements, and actual physical build steps. Each model is also asked to review and critique its own work to see how well it can reason about mistakes and improvements.