I think it's important that we call out its shortcomings, when common bugs plague the system (think llms with hallucinations) or make it act in unison compounding risk ( think the algo Aladin used by 90% of trading firms on Wall Street, causing similar assets pics and collusion while keeping the firms hands clean as they never communicated with each other, they just used a predicitve algo that chose the same positions, which should still count as collusion but i guess not when congress cant even figure out a web browser). There are certain repetitive acts in which ai will be wildly successful, but when it comes to enforcement we can't just rely on ai as it can be exploited. Using it to cut down on creative work loads is helpful and allows for creating deeper art. Using ai to write out boiler plate code so a developer can focus on implementing business logic and security will vastly improve productivity. Where as using it for creating test scenarios will be futile as it only builds off its training data and new edge cases may never be caught. We need to define regulations on where AI can be used in commercial applications, and we need to do it ASAP.