This seems like making SQL developers dumber. I dont understand why would I write all that comment and copy and paste the table definition when I can just write the whole query a lot quicker? 

As for the junior developer I fear they would start relying on this instead of learning the actual query language.

some might think that this is making developers dumber, but when I first started learning SQL back in 2000, they had a built in query builder which now is used to build views (select new view to see what I mean).  I didn't know the first thing about SQL and was just given the manual and MSDN disks and told I was now in charge of report queries, to learn it and fix a bug in a report.  I was fortunately shown how to use the old query designer (right click on a view folder and choose new view to see it) and I used this to construct queries for the first year.  It gave an easy way to pick tables, and showed me what the various types of joins looked like, how to filter data and how to use aliases.  I slowly used it less and less once I knew how the code worked and it became faster just to write the code myself as I got comfortable.  I think generative AI will be something like that on steroids for new developers.  Allowing them to quickly create and experiment with custom code without having to do hours or days of research through expired online help articles and should be a great boon to teach them how to code projects they've never tried before.

I don't know why a developer would use this kind of tool, just to get dumber. 

NomanTee I completelly disagree with you. I don't think you will write SQL queries quicker than the suggestions Copilot gives you.

I had to recently create some tables using some inner joins from multiple tables that used a lot of data, and the keys were really annoying, with complicated names and all... Doing it maunally I was risking running into typos, writing everything maually, which would take a time I did'nt want to waste.

Using Copilot, I literally copied pasted a documentation that I created some time ago using UML, and it did the entire thing by itself, with no errors whatsoever.

What's even best, using Copilot does not make you dumber, instead, it makes you more agile, and even give you some insights on different approachs to a same business logic. Why limit your capacitiy when you can be even better?

Hi there,

-- I have table with two column ID and datetime can you make co pilot produce this code?
--The query grabs the first time for each ID (the first SELECT, before the UNION ALL)
--and then recurses, finding the next row for each ID which is at least 15 minutes appart.
WITH times AS (
SELECT e.ID, MIN(e.DTTM) AS DTTM
FROM example e
GROUP BY e.ID

UNION ALL

SELECT r.ID, r.DTTM
FROM (
SELECT e.ID, e.DTTM,
ROW_NUMBER() OVER (PARTITION BY e.ID ORDER BY e.DTTM) AS rn
FROM example e
INNER JOIN times t
ON e.ID = t.ID AND e.DTTM >= DATEADD(mi, 15, t.DTTM)
) r
WHERE r.rn = 1
)
SELECT top 100 percent ID,DTTM
FROM times
ORDER BY ID, DTTM

I think we the developer need to learn how to ask good questions in order to get right answers. This is classic (for me) example I stuck with the question.

It appears that copilot is acting like a 4GL (4th generation language). And like all 4GLs the instructions are more related to human language but complex instructions are difficult to convey. Think of trying to conveying in English a recursive query involving a dozen or more joins, unions, and subsets (and make that query efficient). What would really be beneficial is a self tuning database, one that would optimize performance based on past queries.  

Late to the party, but…

I spend most of my time in sql writing random queries. At most a few times a year I add a new table, alter a table, or write a sproc. When I do, I often have to stop by W3schools to refresh my memory on some bit of syntax.

my hope is copilot will speed this up. The suggestion may not be perfect, but it’ll get me closer to what I need without having to hit the browser.

I’ve tried the vscode mssql plugin over using SSMS.. I need to spend more time in it next time I have a bunch of sql work to do… the biggest pain point was not being able to run highlighted code. Maybe this has been fixed, or maybe I just need to get in the habit of commenting out queries I’m not interested in executing this run!

I ended up here to see if Microsoft has anything similar to PlanetScale’s Schema Recommendations. I look forward to Microsoft taking some inspiration from it 😉 

The fact that I need to clutter my editor with table definitions in order for copilot to be able to get the context it needs is frankly ridiculous. What if I am working on a very complex database that requires a dozen joins for many queries. That is not unusual where I work. And it would be quite time consuming to extract all of those DDL statements.