[Video] A Very Special Episode of Office Hours

When we were kids, we knew we were in for something heavy whenever a TV series began an episode with a title card explaining, “This is a very special episode of ___.” It meant they were gonna discuss something heavy, like cancer or drugs or loss.

Well, in this Very Special Episode of Office Hours, we’re going to talk about what’s happening to tech jobs in the wake of AI.

Before you watch, I want you to read the post “What I believe about the future of software development” by Thorsten Ball. It’ll be shocking for some of you, but read it with an open mind.

Then, I had AI come up with a list of the kinds of questions you would ask me, and I hand-picked the ones that I thought were the most important and relevant. I don’t use AI-generated questions here often, but I wanted to get this episode out quickly without waiting for y’all to come up with questions, because I think this topic is really important.

Here’s what we covered:

• 00:00 Start
• 03:56 Index Cobbler: Thorsten compares human coders to Italian shoemakers. Is hand-tuning an index or query destined to become a luxury craft, and if so, who will still pay for it?
• 06:53 DROP TABLE Optimist: What permissions, approvals, and blast-radius limits would you require before letting an AI agent make changes directly in a production database?
• 08:38 Second-League Sam: If people without enough tokens are temporarily pushed into a “second league,” what career advice would you give a DBA whose employer will not pay for capable AI tools?
• 11:52 Legacy System Whisperer: The article predicts a generation-long replacement of old software. Does that give SQL Server DBAs a long career runway, or trap them maintaining systems nobody wants to understand?
• 13:26 SSMS Widow: If the terminal and CLI tools are dying, what happens to SSMS, sqlcmd, and Azure Data Studio? Is Management Studio going to turn into a chat window with a query plan hidden somewhere behind it?
• 16:13 The One Percent: Thorsten says performance optimization only matters for the 1% of software that actually has users. Isn’t that 1% exactly the stuff that lands on my SQL Server? Does this article make DBAs more valuable or less?
• 17:21 Compiled But Wrong: He says 900 lines of AI-written C compiled and ran first try, so tests are pointless. How many AI-generated queries have you seen that compiled fine, ran fine, and returned the wrong rows? Isn’t SQL the poster child for ‘it ran’ not meaning ‘it’s right’?
• 20:41 ACID Reflux: How do you reconcile probabilistic AI-generated software with the deterministic guarantees we expect from transactions, constraints, and financial data?
• 21:50 Query Store Historian: If agents constantly rewrite queries and database code, how should we baseline performance and prove which generated change caused a regression?
• 23:33 Meat Proxy DBA: The article says engineers who merely feed tickets to agents will lose value. What work keeps a DBA from becoming a “meat proxy” for AI-generated scripts?
• 25:30 Leading Comma Karen: Thorsten says code style rules like 80 columns exist for humans and agents don’t care. Does that mean the T-SQL formatting wars are finally over? Can I stop arguing about leading commas and just let the robot pick?
• 27:40 Schema Sommelier: If nobody designs schemas by hand anymore, who teaches the model what a good one looks like? Won’t it just learn from every terrible EAV table and GUID clustered key on Stack Overflow?
• 29:20 Review Bombed: Thorsten Ball says code review is dead because humans can’t find bugs in AI-generated code fast enough. Should I stop reviewing the T-SQL my developers’ agents generate, or is the database the one place where somebody still needs to read the query before it hits production?
• 31:32 Wrong-Thing Wrangler: If most future bugs are really “you asked for the wrong thing” bugs, are DBAs about to spend more time validating requirements and less time tuning queries?
• 33:31 sp_BlitzAutopilot: sp_Blitz and sp_BlitzIndex have AI features now. Given what Thorsten says about nobody reading output anymore, is the next step letting the scripts just fix things themselves? Would you ever ship that?

If you’ve got other questions about this topic – or others! – you can ask ’em over at Pollgab.