For me, it's CTE's. I find it amazing to complete a calculation with clear intermediate steps, and goes a long way towards convincing people to use SQL rather than Excel to perform calculations on large tables of data.
What construct do you like using on a daily basis?
I believe recursive CTEs are pretty cool for tree traversal. Anytime you've got a table with a foreign key on its own primary key they might come in handy.
I was skeptical of CTEs for a long time. I just used subqueries when I could in T SQL, and then I got a new job and my new company used Postgres. In the adaptation process I took a new look at CTEs and became a convert - it's just nicer and easier to read the intermediate step than as a subquery
Those work, but require a lot of careful structuring to get right, and can be a pain to debug. With a CTE, you can just call on the intermediate steps to trace down problems.
You can create a functional enum view by just assigning enums as the column names and storing a single row of the int (or whatever enum) representation.
Then use that view in a cross join. You can (almost) eliminate magic numbers entirely and makes the code much more human legible.
Example
CREATE VIEW AS enum.OrderType
SELECT
CAST (1 as 'New'),
CAST (2 as 'Pending'),
CAST (3 as 'Shipped')
GO
-- Assuming a table with OrderId and OrderTypeId
SELECT o.OrderId
FROM dbo.Orders AS o
CROSS JOIN enum.OrderType AS ot
WHERE o.OrderTypeId = ot.[Pending]
-- Only returns orders where TypeId = 2, no need to know what Id that is or for anyone else to in the future either.
This trick works even on large datasets with a lot of complex joins. Getting the status name itself to return is a bit more of a chore though.
CTEs can be useful, particularly in PostgreSQL, where there are writable CTEs, but a lot of the time, I prefer using temp tables over CTEs, as they often perform better for larger datasets. I think one of my favorite constructs is window functions. I've found many uses for them, over the years.