That's true for driverless cars but there's many areas where you don't need 100% to be useful. Stable diffusion and LLMs are 90% there but still very useful and cool.
Most of those areas don't normally have a lot of value or you're marginally improving on existing recommendation systems. It's a coin toss on how much of an improvement this wave of LLM powered ML systems will have over the previous wave of "enterprise knowledge collation" systems and if that will be enough for people to buy them.