In a previous life I worked for a company that was developing real-time ray tracing products. The founders had this magical algorithm but the catch was that dedicated ray tracing hardware was almost never successful because by the time the dedicated hardware made it to market general purpose processors had caught up.

However, what it seemed like to me was that the founders had developed a blazing fast algorithm that cut a ton of corners. Each time they'd fix an edge case the product got slower. Regardless they were moderately successful and might still be around.

And then there was the time I accidentally nuked all of our internal infrastructure in the middle of a product release demo.