This is a subject that evolves every day, so it's hard to give an accurate answer.
To make it harder, there is no "The AI". There are many Generative AI agents. I suppose you're asking about their capabilities.
An easier question would be what they can not do. They are often regulated not to provide information that is deemed dangerous, such as building bombs or chemical weapons.
Many AI coding agents claim that they can create the software you need with a few prompts. As a software engineer, I would take this with caution. Yes, most of the time, the generative AI tools can create you some kind of a software that runs, but they do not necessarily do what you need. So you'd need to review the code, change it to meet your business needs, and test if it really does what it's supposed to do correctly. Some companies even say that their AI Agents can do the testing.
At this stage, AI-generated code can be used as a scaffolding, but no one should build mission-critical software purely on AI code. I wouldn't want to fly on a plane programmed by AI only, or on a spaceship; or get a medical procedure overlooked by AI only, or have AI manage my finances without 0 human review.
If you get AI to generate a game for you, the worst thing that would happen is that the game is not fun. But if your plane crashes because the AI messed up the altitude calculation, that's unacceptable.
That's why I say that AI-generated code needs to be supervised (and not supervised by another AI).
Go, search the Web, there's plenty to learn about AI and the state of art changes every day.