Judging from the rate of improvement in AIs it will happen gradually over the next few years.
Comparing AIs of a few years ago to the ones today there is a good chance they will improve enormously.
Some humans will always be needed for handholding, but not as many as employed today to churn out run of the mill code.
Sure that's a theory that sometimes plays out, but similarly one also sees almost exponential growth or improvement, yet followed by a plateau for a long time until the next breakthrough.