Introduction
AI-assisted data development is reshaping the way SQL is traditionally written. Today’s mainstream approaches can be broadly divide...
For further actions, you may consider blocking this person and/or reporting abuse
What I like here is that AI isn’t being treated as the final authority. It translates a fuzzy business requirement into something structured, then a deterministic layer takes over and humans still validate the business meaning. That feels much closer to how AI-assisted development should work in production.
Case 3 is a great example: the first solution could be syntactically valid and still be logically wrong because LineID was the wrong relationship. No compiler can understand that business assumption for us.
I’d be curious about one thing as SQL complexity grows: where does SQLazy’s abstraction ceiling appear? Recursive CTEs, complex window logic, database-specific behavior, query-plan-sensitive optimizations, etc. At some point an intermediate language can become complex enough that we’ve effectively created another SQL dialect.
Still, I really like the core idea here: use AI to structure the uncertainty, but keep validation and execution deterministic. Nice work.
Hi,Mustafa , t's free. Why don't you give it a try?
Haha, fair point 😄 You got me.
I’ll give it a try. I’m especially curious to see how far the abstraction holds once the queries move beyond straightforward transformations into recursive CTEs, heavy window-function usage, and execution-plan-sensitive workloads.
That’s probably the best way to answer my own question anyway — break it with real workloads. 😄
Thanks, Judy!
Yes, you are really great! Mustafa~
Well… I took your advice and actually tried it 😄
My first impression is that SQLazy makes much more sense once you use it than when you only read about it.
What stood out to me most was the intermediate-step model. Being able to inspect the transformation step by step makes complex data logic surprisingly easy to reason about. It also creates a nice boundary between AI-generated reasoning and deterministic execution.
I deliberately tried to approach it with the question I mentioned earlier: “When does the abstraction start getting in the way?”
So far, I haven’t hit that wall as quickly as I expected. 😄
I still want to push it with more difficult cases — especially complex window logic, database-specific behavior, and performance-sensitive queries. That’s where I think the really interesting test begins.
So yes, Judy… you were right. Trying it was definitely better than just talking about it. 😄
Hi Mustafa，You are amazing! Looking forward to your sharing of usage cases.