A code-completion agent is extremely far removed from the quality of the output of a frontier coding model. Try Claude or similar and ask it to generate code specific-well defined tedious tasks. You may be surprised!