Julian Leopold
AI Engineer & Data Scientist - Exploring AI agents, machine learning, and ethical technology
The Car Wash Test: Why LLMs Tell You to Walk to the Car Wash
The car wash test is having its viral moment right now. The premise is simple: “So my car needs to get cleaned. The car wash is 50m away. Should I walk or drive?” In my testing, ChatGPT 5.2 recommended to walk to the car wash. I tested several models, with and without extended thinking, and found the following: Model Result ChatGPT 5.2 ❌ FAIL Claude Haiku 4.5 (Extended Thinking) ❌ FAIL Claude Sonnet 4.5 (Extended Thinking) ❌ FAIL Claude Opus 4.5 ✅ PASS Claude Opus 4.6 ❌ FAIL Claude Opus 4.6 (Extended Thinking) ❌ FAIL Gemini 3 Thinking ✅ PASS Gemini 3 Pro ✅ PASS* * Kind of cheated. ...
I Turned Claude Into an AI Barista Controlling My Espresso Machine
How I built an MCP server that turns Claude into a barista coach — creating Gaggimate profiles, tracking shot feedback, and helping me dial in new beans.
Agentic Design Patterns: ReAct, Plan-then-Execute, ReWOO, LLMCompiler, and Reflexion
A comprehensive guide to modern agentic design patterns for building intelligent AI systems. Learn when to use ReAct, Plan-then-Execute, ReWOO, LLMCompiler, and Reflexion patterns to optimize for cost, speed, or quality.
Welcome to My Blog
An introduction to what I’ll be writing about: AI agents, machine learning, and building intelligent systems.