The word every AI vendor puts in front of a dealer is "competency." A federal regulator now writing the test for what that word means in a self-driving car has just conceded there wasn't a test yet. The gap between the word and the test is where every retail-AI decision has been getting made too.

On July 16, NHTSA Administrator Jonathan Morrison told Bloomberg that the agency will start drafting objective "behavioral competency" standards for autonomous vehicles, backed by tests a manufacturer can pass or fail. Morrison's own phrasing: so a maker "can know with certainty whether or not the vehicle meets the requirements." The move follows a June recall of thousands of Waymo vehicles after cars drove through freeway construction zones at speed, and a running scrutiny of the fleet's failure to slow or stop for school buses.

That is a plain admission worth naming. Every autonomous-driving demo you have watched in the last seven years ran on the vendor's own test data. So has every retail-AI demo. When a service AI vendor claims 92% first-call resolution, resolution measured against what queue, whose calls, which advisors? When an F&I compliance AI says it catches disclosure gaps, gaps against which state's requirement, on whose deal jacket? The test lives inside the vendor's product team, not inside the store.

What NHTSA is proposing to write is a test a third party can run. That is a different product than a test that grades the demo. It is also the product that has been missing from every AI conversation dealers have been having for three years. If "competent" is not defined against something the buyer can measure, "competent" is whatever the vendor says it is.

The operational read is not "wait for the regulator." Regulators will write tests for the failure modes they can see, cars that do not stop for school buses. They will not write the test for the failure modes that live in your store. A phone AI that lets a returning customer sound like a new one. A CRM AI that misroutes trade-in inquiries during service hours. A schedule bot that books a vehicle in for the wrong tech skill. Nobody at NHTSA is writing that test. You are.

That is what an operating standard is. It also means the next vendor demo that lands "AI competency" in your conference room has a question waiting for it. Not "does it work." "Against what test." If the answer is a scenario the vendor built for the demo, you are watching a product get graded by its own product manager. The dealers who work out what the right test looks like will work it out with other dealers, not with vendors. That conversation does not happen on a demo call.

Robotaxi passengers behaving like passengers — a Bloomberg feature catalogued riders falling asleep, spilling food, and prompting 911 calls from unresponsive robotaxis; the story was widely shared but does not move a franchise store's P&L in the next quarter.

Q1 2026 set a record for TCPA class action filings — 220 in March alone.

FEATURED · SEPTEMBER 24 · CAMBRIDGE

The Automotive AI Summit is September 24 at the MIT Museum in Cambridge. It is built for dealer principals and operators, not keynote talent: no pay-to-speak slots, no vendor floor. Confirmed for the day: Yuriy Demidko, whose Fox Motors data work anchored last week's issue and who is now Chief Technology and Product Officer at Clarivoy; Nolan Armpriester, general manager of Twin Pine Ford of Hamburg, who built the country's number-one volume mobile service dealer with over 35,000 appointments in 2024; and Joe Castelino of American Motors Group, whose line is that dealers have a growth problem, not a traffic problem. Whichever test you are writing for the AI running through your store, the operators writing theirs will be in the room. Reserve your seat.