Over the past two decades, the volume of electronics and software in cars have grown tremendously. But this growth has also resulted in hardware and software architectures that are proving to be a bottleneck for further innovation and efficient design flows, especially when implementing compute-intensive functions necessary for modern autonomous features. For example, centralized architectures that are driven by the use of more powerful processors result in higher sensor-to-actuator delays. Similarly, timing uncertainties increase as signal-based in-vehicle communication is being replaced by more dynamic service-oriented communication architectures. Finally, the increasing volume of software running on powerful multicore ECUs is making timing analysis, including WCET estimation, to be very complex. As a result, timing estimates, when safe, are very pessimistic, which makes efficient implementations to be difficult. In this position paper, we outline some of these emerging challenges and discuss potential solutions.