Yep - well said George.
I've been working on a project at work to convert 100's of TSQL stored procs into Views - but Snowflake SQL views.
(first I convert the sproc into a view in tsql, then go from there to convert into Snowflake).
I was brand new to Snowflake a few months ago, so this is a big deal to me, and heavily reliant on ChatGPT to help me figure out how to replace my tsql favorite stuff - outer apply's for example - into something snowflake likes.
The other day I got frustrated at ChatGPT but for the wrong reason. I had kept telling it "I'm getting subquery cannot be evaluated error", and hoping ChatGPT could figure out why. It kept sending me down rabbit holes that didn't work. Then later I realized the real problem in the query had been with a LEFT JOIN LATERAL - not a subquery. At first I was frustrated, but then I thought to myself Wait, wasn't it MY responsibility to isolate the part of the code that was causing the problem? As soon as I isolated it and told ChatGPT the part of the code that was really the problem, it solved the problem quickly. Up to that point, I had been lazy, basically, pasting in my whole View and telling it to find the problem. I could have easily isolated the problematic portion quickly and told ChatGPT to focus on that, but I hadn't.
So using AI still requires a LOT of smarts and thought into what you input. I have a feeling developers won't be replaced so much as become more productive - unless they make the dumb mistake I did, which is to keep pasting in the whole view and asking what the problem was, instead of commenting out chunks so I could isolate it myself.
(Poor ChatGPT tried at least 10 solutions on a part that wasn't even the problem until I isolated it properly).
