Why AI Can't Replace Your Judgment Yet | JD Margulici (Cruise)

**JD Margulici is a technologist with more than twenty years of experience** building information technology for transportation, industrial automation, and autonomous driving. He cofounded Vehicle Data Science in 2013, pioneering the use of connected fleet data to build high-definition maps, and later served as CTO at Infinite Uptime and a senior staff engineer at Cruise. Before that, he trained in mathematics and physics at Ecole Polytechnique in Paris, then earned a master's in transportation engineering and an MBA at UC Berkeley. In our conversation, we discuss: 1. How AI has shifted from an autocomplete assistant to the default first move in his day to day coding and design work, and what that shift costs in terms of owning your own knowledge 2. Why raw intuition and taste, not access to tools, is what now separates good work from bad, since everyone has access to the same AI 3. The real bottleneck in software engineering: not writing code, which is under 20 percent of the job, but interpersonal communication, translating vision into requirements into architecture into code 4. A cautionary story about two AI-written memos on his team's wiki that sounded confident but didn't hold up, and why he chose not to argue with an AI's reasoning 5. Physical AI and the "floating brain in a tank" problem: why language-only intelligence still lacks a toddler's intuition for how the physical world works, and what robotics could change 6. The structured to unstructured to structured pattern that makes AI agents actually useful, and why he's still unsure what to tell his own kids about their future **Where to find JD:** LinkedIn: https://www.linkedin.com/in/jdmargulici/ **Where to find Vlad:** Website: vladmoskovski.com LinkedIn: linkedin.com/in/vladmoskovski **Free Somatic Intelligence Toolkit for Leaders** AI gives everyone data. The leaders who stand out also listen to their body's intelligence. Download the free toolkit here: vladmoskovski.com/handbook-landing-page **Timestamps:** 00:00 Introduction, and being the first ever podcast guest 00:46 How the day to day of coding and design has shifted with AI in the past year 01:54 Democratized tools, and what actually separates good work from bad now 04:11 Why interpersonal communication, not code, is software engineering's real bottleneck 07:24 The AI-written memos on the wiki that didn't hold up 09:34 AI slop, and the flood of easy to generate, hard to evaluate content 13:07 Guardrails, and hoping the growing pains are just transitional 14:29 Physical AI, digital twins, and managing real-world infrastructure 16:58 The "floating brain in a tank" without senses 19:30 Representations, notation, and why code is often the most precise language 23:11 The structured to unstructured pattern that makes AI agents work 26:10 Closing thoughts: feeling clueless, parenting, and what to tell his kids about the future