When you ask GPT 3 to produce some code it's a bit hit and miss but it's not to bad. This is enough to start to replace programmers by the end of the year. The mistake people make is that they assume it has to be perfect at code and able to do more than just a few lines of code in a competent to replace the programmers after all you want the code to work and function correctly right If you tried to get GPT-3 to code a small app it likely would just program something which didn't even start properly. people producing points of code in google are corrected by their team manager if they make a mistake and knitting the code together and point testing are other tasks. GPT-3 can do what the person writing points of code for the team manager can do and can learn from being corrected efficiently enough.
The more GPT 3 is corrected the better it would get at the task much the same as a human.