Stop Celebrating Robotaxi Maps. Watch the Weather Days.

Every few months a new city lights up on a coverage map and the narrative resets: autonomous driving is here. Maps are good PR. They are weak evidence.

So basically: geographic expansion is a milestone, not a product. What matters is how often the system drives without help, in which conditions, at what cost per mile.

Coverage ≠ competence

A geofence that works downtown at noon on a clear day is not the same as a network that handles:

• Construction that appeared overnight
• Heavy rain at dusk
• Aggressive merging on multi-lane arterials
• Edge cases that don’t show up in the training set until week three of operations

Operators know this. Press releases often don’t emphasize it.

The metrics worth asking for

If you’re evaluating an AV operator — as a rider, investor, city, or journalist — ask for the boring numbers:

• Miles between disengagements (and how “disengagement” is defined)
• Share of trips completed without remote or in-car assistance
• Performance by weather and time of day, not just fleet averages
• Cost per revenue mile after operations, insurance, and support staff

Without those, “we’re in six cities” is a slide, not a thesis.

Cities are not interchangeable

ODD (operational design domain) is the quiet villain of every AV story. A system validated in Phoenix sun does not automatically transfer to Boston winters or dense European cores. Mapping, local driver culture, regulatory posture, and road geometry all move the needle.

That’s why expansion often looks cautious even when the tech demos look bold. Caution is the product working as intended.

The filter’s take

Autonomous vehicles are past the “impossible” phase and deep into the “hard engineering and operations” phase. That phase is less viral and more important. Watch utilization and reliability curves, not just pin drops on a map.