Recently had to deal with some lengthy SQL query, which has window functions built on top of previous window function results, took a lot of effort to track down the problematic part. For this task AI wasn't really could not help at all at some point, but later on I realized I was using models that have smaller context window limitation. Maybe with more powerful model it might've helped.
I guess using proper model would be a helpful tip for myself down the road, but maybe soon computational costs just get much cheaper that everyone is supplying large context window.