MIT's JARVIS Challenge (Jet-engine AI Research and Validation Intensive Sprint) had undergraduate students design, fabricate, and test a small jet engine in four weeks using AI as their primary engineering partner. Teams used LLMs like Claude and ChatGPT for trade studies, design comparisons, and knowledge gaps, but encountered hallucinations, sycophancy, and lack of physical intuition. The winning team (811 Crew) was more skeptical of AI and relied on engineering fundamentals, while the fastest-moving team leaned heavily on AI. Key findings: engineering experience amplifies AI value, manufacturing remains the rate-limiting step, vendor relationships can't be replaced by AI searches, and human judgment is irreplaceable for safety-critical physical systems. The challenge suggests AI can compress design-build-test cycles dramatically, but only when engineers have the expertise to direct and verify AI outputs.
