Some of these arguments are intellectually interesting, but I struggle to see strong real-world evidence behind them.
In practice, meaningful results with AI still seem to depend on people who understand data modeling, semantics, and system constraints. AI amplifies that knowledge; it doesn’t replace it.
I’d genuinely be curious to see concrete examples where abandoning modeling entirely has produced better, more reliable outcomes at scale.
The idea that "data modeling is dead" is ridiculous, I agree. It's more important than ever.
But I also disagree that we've figured it all out with "star schema", "Kimball" or "3NF". I think modeling for the usage is hyper important, and in some cases the user is AI, where methods like "OBT" (ish, literally one table is silly) really do work better.
I'm doing a fair bit of this work in the real world, so I'm curious to read Joe's arguments.
It’s interesting that this xyz is dead message follows such a cause and effect path. “Because of this new tech xyz now must die”. It’s like the people crafting these arguments never learned that coexistence and compliments are a thing.
It’s now just clickbait for likes. Just don’t say excel is dead, every business leader director level and up will have a heart attack. “Where’s my PowerPoint of excels?”
Not only is data modelling more important when providing better context for ai driven data products, but it's also making the maintenance and documentation of data models so much quicker and easier. I'd argue we'll actually see better adoption because of this.
I think what bothers me most about these arguments is that there are threads of insight in them - but instead of pausing to ask the question of “How might things change is machines get good at XYZ?” or “What challenges might arise if we try to use AI in this way (cost, accuracy, latency, scale, etc)?” or “What assumptions am I making here?” —instead of that they jump straight to hand waving across all the complexity and jumping straight to the “And therefore this thing is dead.”
It’s lazy, extreme, and ultimately - probably wrong. I feel like people do this either as rage bait because algorithms reward it over nuance. Or maybe out of fear themselves. Either way I hate it. It’s not constructive and doesn’t help us.
I want conversations that help us think deeply about what’s changing and pivot based on our best understanding of where things might be heading.
Hello, here is a new Data Engineer apprentice. I recently read a passage from a book where the author or authors emphasize the need to have a solid data foundation even for later working with AI, ML, LLM models. Hoy can often see comments and posts trying to convince you that AI will take over a largo parte of your job responsibilities and that it will no longer be necessary yo learn skills that are fundamental for specific roles. The sad thing is that for someone like me who doesn't have much knowledge and experience un IT, it creates confusion and discouragement...
A slightly tangent to this but it always made me laugh that there was a huge amount of noise and excitement behind NO SQL databases a few decades ago. But low and behold it wasn’t long before we had SQL being used against these NO SQL databases as the go to mode for interrogating them.
I think we go through the same type of pattern again and again with new technology and forget that the old techniques aren’t redundant, in fact they can not only complement but can enhance new technologies capabilities.
It’s also I think part of the human DNA to assume that what’s new must be better and what’s old isn’t needed anymore. I remember back in my data architecture days having to review an analysis of technical options for a new database solution we where building that had been put together by a few of the junior members of my team. I remember them comparing Mongo DB and Oracle and writing an outrageously biased comparison of the two technologies. Based upon their analysis there could only be one choice. They had completely overlooked the Oracle RDBMS capabilities because in my opinion they wanted to work with Mongo DB rather than Oracle, as they perceived it as being old hat.
Every one of these arguments for data modelling is dead, falls flat when you optimise for cost. Only companies or teams with enough money can say this. The rest of the companies with sensible budgets will always know where to start.
"Schema-on-read 2.0" made me shudder. A friend of mine started using the phrase (to describe data I must clarify) "It should always hurt more going in than it does coming out"
As usual a thought provoking post. I started philosophising in my head - if people give all their knowledge work to AI, what is the point of existing. It got deep, quickly.
While I think it’s obvious you don’t believe that data modeling or the value in learning things like SQL are dead, why is that the case Joe?
I’m going to assume it’s because one needs to understand what they’re doing in order to seperate “good” from “bad” or “right” from “wrong”. And in doing so, one gains some kind of efficiency or advantage (economic or otherwise)?
Some of these arguments are intellectually interesting, but I struggle to see strong real-world evidence behind them.
In practice, meaningful results with AI still seem to depend on people who understand data modeling, semantics, and system constraints. AI amplifies that knowledge; it doesn’t replace it.
I’d genuinely be curious to see concrete examples where abandoning modeling entirely has produced better, more reliable outcomes at scale.
The idea that "data modeling is dead" is ridiculous, I agree. It's more important than ever.
But I also disagree that we've figured it all out with "star schema", "Kimball" or "3NF". I think modeling for the usage is hyper important, and in some cases the user is AI, where methods like "OBT" (ish, literally one table is silly) really do work better.
I'm doing a fair bit of this work in the real world, so I'm curious to read Joe's arguments.
Same
As I’ve been writing and doing in practice, pick what works for your situation
It’s interesting that this xyz is dead message follows such a cause and effect path. “Because of this new tech xyz now must die”. It’s like the people crafting these arguments never learned that coexistence and compliments are a thing.
It’s now just clickbait for likes. Just don’t say excel is dead, every business leader director level and up will have a heart attack. “Where’s my PowerPoint of excels?”
Bingo. It’s a weird zero sum argument
Not only is data modelling more important when providing better context for ai driven data products, but it's also making the maintenance and documentation of data models so much quicker and easier. I'd argue we'll actually see better adoption because of this.
I think what bothers me most about these arguments is that there are threads of insight in them - but instead of pausing to ask the question of “How might things change is machines get good at XYZ?” or “What challenges might arise if we try to use AI in this way (cost, accuracy, latency, scale, etc)?” or “What assumptions am I making here?” —instead of that they jump straight to hand waving across all the complexity and jumping straight to the “And therefore this thing is dead.”
It’s lazy, extreme, and ultimately - probably wrong. I feel like people do this either as rage bait because algorithms reward it over nuance. Or maybe out of fear themselves. Either way I hate it. It’s not constructive and doesn’t help us.
I want conversations that help us think deeply about what’s changing and pivot based on our best understanding of where things might be heading.
Bingo
Hello, here is a new Data Engineer apprentice. I recently read a passage from a book where the author or authors emphasize the need to have a solid data foundation even for later working with AI, ML, LLM models. Hoy can often see comments and posts trying to convince you that AI will take over a largo parte of your job responsibilities and that it will no longer be necessary yo learn skills that are fundamental for specific roles. The sad thing is that for someone like me who doesn't have much knowledge and experience un IT, it creates confusion and discouragement...
Very understandable
A slightly tangent to this but it always made me laugh that there was a huge amount of noise and excitement behind NO SQL databases a few decades ago. But low and behold it wasn’t long before we had SQL being used against these NO SQL databases as the go to mode for interrogating them.
I think we go through the same type of pattern again and again with new technology and forget that the old techniques aren’t redundant, in fact they can not only complement but can enhance new technologies capabilities.
It’s also I think part of the human DNA to assume that what’s new must be better and what’s old isn’t needed anymore. I remember back in my data architecture days having to review an analysis of technical options for a new database solution we where building that had been put together by a few of the junior members of my team. I remember them comparing Mongo DB and Oracle and writing an outrageously biased comparison of the two technologies. Based upon their analysis there could only be one choice. They had completely overlooked the Oracle RDBMS capabilities because in my opinion they wanted to work with Mongo DB rather than Oracle, as they perceived it as being old hat.
Every one of these arguments for data modelling is dead, falls flat when you optimise for cost. Only companies or teams with enough money can say this. The rest of the companies with sensible budgets will always know where to start.
"Schema-on-read 2.0" made me shudder. A friend of mine started using the phrase (to describe data I must clarify) "It should always hurt more going in than it does coming out"
As usual a thought provoking post. I started philosophising in my head - if people give all their knowledge work to AI, what is the point of existing. It got deep, quickly.
While I think it’s obvious you don’t believe that data modeling or the value in learning things like SQL are dead, why is that the case Joe?
I’m going to assume it’s because one needs to understand what they’re doing in order to seperate “good” from “bad” or “right” from “wrong”. And in doing so, one gains some kind of efficiency or advantage (economic or otherwise)?