They had a huge smalltalk program that created embedded C code (AUTOSAR/MISRA if I remember correctly). Tests only ran manually and the coverage wasn't particularly high. I don't recall any automations that would have stopped you from saving new code that breaks tests to the smalltalk image.
Even though it 'worked', I wouldn't say this whole system excelled at what it did. Lots of manual QA.
So technically it worked, but it certainly was in poor shape from an automation perspective. It was my job as the mostly unsupervised intern to fix that. Granted, that was 10 years ago, but talking to some of my friends around Stuttgart, things move at a glacial pace :)