If it helps, it doesn’t code all that well. You have to spend hours, days even, putting guard rails around how it will write code.
Then, agents check all of the code that is written, which misses a lot, and then tools to monitor / report on the code it deploys.
And, that’s ideal apparently.