Different pieces of software have different quality. Websites are usually on the crappy end of the scale. Central parts of operating systems are towards the opposite end. Also, many commercial products are developed with little testing. But there are methodologies for better testing, even mathematical proofs of correctness. Those are usually not used in commercial development, because they require some time and qualification, and companies prefer to hire cheap coders and have the product soon, even if it is full of bugs. And generally, because software companies are usually managed Dilbert-style.

However, it is possible to have mathematical proofs about algorithm correctness (any decent university teaches these methods as parts of informatics), so in these debates it is usually assumed that people who would develop an AI would use these methods.

To a person who knows this, your comment sounds a bit like: "my childhood toy broke easily, therefore it is impossible to ever build a railway that would not fall apart below the weight of a train".