Diagnosing a grinding engine noise demands a mechanic's hands, not an AI's text.

Every grinding noise has a different cause — worn brake pads, failing wheel bearings, low oil, transmission issues — and diagnosis depends entirely on the specific vehicle, its history, and real-time sensory data. No two instances are structurally the same, making repeatable automation impossible.

The success criterion is a definitive, accurate safety judgment about a specific physical system. That bar is extremely crisp and high-stakes — vague or probabilistic output is not acceptable when the question is 'will this car fail on the highway tomorrow.'

An AI agent has no access to the actual vehicle, cannot hear the noise, cannot read OBD-II codes without a connected tool, and cannot physically inspect components. The core diagnostic data is entirely unavailable to any remote agent.

A false 'safe to drive' verdict could lead to a catastrophic mechanical failure at highway speed, risking the driver's life and others on the road. This is an irreversible, potentially fatal error — the highest possible cost category.

A trained mechanic uses tactile feedback, auditory pattern recognition, visual inspection, and years of hands-on experience to diagnose engine problems. This is precisely the kind of embodied, contextual expertise that AI fundamentally cannot replicate remotely.