The rising complexity of automotive software makes it increasingly difficult to develop the software with high quality in short time. Especially the late detection of early errors, such as requirement inconsistencies and ambiguities, often causes costly iterations. We address this problem with a new requirements specification and analysis technique based on executable scenarios and automated testing. The technique is based on the Scenario Modeling Language for Kotlin (SMLK), a Kotlin based framework that supports the modeling/programming of behavior as loosely coupled scenarios, which is close to how humans conceive and communicate behavioral requirements. Combined with JUnit, we propose the Test-Driven Scenario Specification (TDSS) process, which introduces agile practices into the early phases of development, significantly reducing the risk of requirement inconsistencies and ambiguities, and, thus, reducing development costs. We overview TDSS with the help of an example from the e-mobility domain, report on lessons learned, and outline open challenges.