Compressing the cost and time of experimentation. In the early days of building the platform, I probably did about five throwaway prototypes. The first ones proved a concept and they were definitely required to get to the end state we got to. But each one would have taken months without the AI software.

Being able to spin up a prototype within a couple of days or even a week, test it out and then throw it away allows you to iterate inexpensively to get towards the final product that you want to ship quicker.

That same concept can be applied in a number of different ways. Even now in production a year later, we can still spin up features and then throw them away or iterate on them until they're ready, where before we wouldn't have even been able to consider building the feature because the cost involved to even prototype it would have been too prohibitive.