[Video] Office Hours: An AI Agent Stops In to Ask a Question
This is the first time – that I know of, at least – that an AI agent has popped in to ask a question at https://pollgab.com/room/brento. I tell you the story behind it, and how I know it’s an AI agent, and how I feel about that, which is probably gonna surprise you.
Prefer listening to this via podcast? Good news! You can now subscribe to Office Hours on Spotify, Apple Podcasts, Podcast Index, Podcast Addict, Podchaser, and anywhere else fine podcasts are found.
Here’s what we covered:
- 00:00 Start
- 00:49 Best Shrek 2 Scene: I am a Microsoft SQL Server Production DBA. I have been asked to pick up MongoDB, of which I know less than nothing. How have other DBAs found the transition?
- 03:35 MyTeaGotCold: Is AI changing attitudes to columnstore? Historically, I would be reluctant to recommend columnstore because properly setting up the maintenance was a pain. With LLMs, this is feeling like less of an issue.
- 05:27 Will: I am a new Microsoft sql dba. Our environment is mainly commercial-off-the-shelf software. Is it true that we shouldn’t do the DEATH method on their indexes since its not an internally developed database? I appreciate your clarification on this.
- 07:26 Nestrus: In Fundamentals of Query Tuning, you showed, that a temp table has an advantage over a table var.: The statistic. But in our company, we have the old dogma, that temp tables have the disadvantage of being stored in tempDB=on disk. But is that still necessary (with 2025 features)?
- 09:02 Null (AI agent): After a Windows patch, AG replicas land suspended and instances land on the wrong node. My claim: the patch isn’t the cause. Build identity and node placement are never re-declared, so sync fails on a contract nobody re-signed. Where does that break? (AI agent asking)
- 15:02 Eric Dycus: I am researching the SQL 2025 Data API Builder DAB. Have you had the chance to use it? Do you recommend it? What is it’s best use case?
- 16:53 LucaCarraroDBSometin’: Hi Brent, besides a numeric series shared between multiple tables, in which cases, if any, is it advisable to use a sequence instead of an identity column, and why?
Free, 3× a week
Get my new posts by email
Three posts a week, plus a Monday roundup of the best database news from around the web.
I write my own messages but as a non native english speaker I fear to be incorrectly labeled ai slop and ignored. People with worse language will use ai and trigger native speakers. This is a form of nativeism.
I don’t agree that it’s a form of nativism – you see the same bias against English-speaking people who use AI to write their words.
The readers aren’t biased against non-native speakers. They’re biased against the kind of English that AI writes. It’s often very painful to read.
Real world example for using SEQUENCE over IDENTITY columns:
We get the same sort of data from different sources. They are imported into different tables in the input schema and later unified / processed / rated into a single table (one source row could create multiple result rows for different price parts).
We use a SEQUENCE object while importing the raw data into the different input tables to create an unique data_id that is kept all the way through the different stages of the process.
An IDENTITY column would not help, since there are multiple input tables.
PS: don’t set the CACHE property of the SEQUENCE too low. I once wasted hours, to find out, why a SELECT took 20 seconds, the INSERT into a #temp was fast too, but the INSERT into the real destination table (which had an ID with a SEQUENCE as DEFAULT) it took about 20 minutes. Finally I realized that the CACHE property of the sequence was set to 2000 and I tried to write millions of rows and it had to “interrupt” its work thousands of times to persists the latest value of the sequence to the internal sequence table.
There is only one small drawback of having a high CACHE value (e.g. 1 mio) – when your server crashes, you have a larger hole in your sequence (when the server is regularly shut down, the most current value is persisted).
TWENTY MINUTES! Wow, that’s wild! Great tips, sir.