Ask HN: How will AI impact the future of CRUD apps?
25 comments
In my experience, CRUD apps are complicated with business requirements and therefore bespoke business logic. AI will definitely assist with getting apps up and running but there will still need to be someone who takes the requirements, turns it into a functional product, updates it, maintains it and owns it.
I doubt it will be business/stakeholder people interacting with the AI. It could potentially be business analysts but I doubt they'll want to. It would be an addition to their current job.
That leaves the software engineers. Maybe a lot of software engineers will turn into _solution_ engineers or _product_ engineers. Their job will be to create the solution/product even if they're not writing code.
I doubt it will be business/stakeholder people interacting with the AI. It could potentially be business analysts but I doubt they'll want to. It would be an addition to their current job.
That leaves the software engineers. Maybe a lot of software engineers will turn into _solution_ engineers or _product_ engineers. Their job will be to create the solution/product even if they're not writing code.
I'm pretty sure I get paid to take everybody's collective best guess at what requirements should be (sometimes with way too many people involved giving their concerns/opinions, or just not being able to get anybody's attention and everybody rubber stamping it/half-assing it), put it in QA/CAT/UAT, let somebody else bang on it, then we all figure out in our effort to try to not miss anything, we missed a bunch of massive design flaws. Then we go back to the beginning or scramble to fix it really fast, and the whole design repeats itself.
At least that's what it feels like I get paid to do...
At least that's what it feels like I get paid to do...
Ha, sounds familiar. I doubt AI will be able to automate that sophisticated process ;)
I suppose AI needs to be able to reason and we are still a long long time away from AGI according to experts.
Btw, what’s the CAT step?
I suppose AI needs to be able to reason and we are still a long long time away from AGI according to experts.
Btw, what’s the CAT step?
client accessible testing/user accessible testing
another environment before prod but not internal only so our “clients” can hit it for testing
another environment before prod but not internal only so our “clients” can hit it for testing
Thanks. Where I've worked we only had Dev > QA > UAT > Prod - so not heard of CAT before.
Often the business requirements are non-sensical, contradicting, or pointless. Someone has to actually figure out what they mean without pissing off steakholders.
Remember, this isn't actually AI, it's machine learning, and the learning part is a misnomer.
Remember, this isn't actually AI, it's machine learning, and the learning part is a misnomer.
> Often the business requirements are non-sensical, contradicting, or pointless. Someone has to actually figure out what they mean without pissing off steakholders.
That reminded me very much of this skit "The Expert". https://www.youtube.com/watch?v=BKorP55Aqvg&ab_channel=Lauri...
That reminded me very much of this skit "The Expert". https://www.youtube.com/watch?v=BKorP55Aqvg&ab_channel=Lauri...
>Their job will be to create the solution/product even if they're not writing code.
Yep.
You can make a good argument that today, any technical profession doesn't have to know as much because of existing software in most any given sector, whether its CAD software with built in stress/CFD analysis for mechanical parts, or frameworks for actual software development that are larger building blocks then pure code.
In the same way, future software engineers will likely be using generative software based on models like GPT that can take plaintext English and translate it into code. There will still be knowledge required of which model to run, what the parameters do, how to tweak the output, e.t.c
Of course, in the further future (although exponentially less time), those engineers will no longer be needed because there will be general models trained on all of their work as well. But really though, at that point, we wouldn't be that far from AGI.
Yep.
You can make a good argument that today, any technical profession doesn't have to know as much because of existing software in most any given sector, whether its CAD software with built in stress/CFD analysis for mechanical parts, or frameworks for actual software development that are larger building blocks then pure code.
In the same way, future software engineers will likely be using generative software based on models like GPT that can take plaintext English and translate it into code. There will still be knowledge required of which model to run, what the parameters do, how to tweak the output, e.t.c
Of course, in the further future (although exponentially less time), those engineers will no longer be needed because there will be general models trained on all of their work as well. But really though, at that point, we wouldn't be that far from AGI.
It won't.
What you've described is a bare bones CRUD app. The little I know about things like AI/LLM is you feed it text so it can learn. If the input is not good then the output is not going to be good either, doesn't matter what it does with the input.
We (as an industry) can't get feature requirements or business logic documented to be interpreted consistently by humans, who understand those problem domains in high fidelity, let alone some computers reading that text. If the translation of requirements to code isn't great, code to LLM to produce new code isn't going to cut it either.
Our industry jokes about all we do is CRUD apps but once an app is mature and beyond simple models, has integrations from a dozen APIs, has customers integrating via APIs, does reporting, needs to guarantee-ish transactions and most importantly is using derived sets of data for billing/invoicing it is much much more than a "CRUD app".
What you've described is a bare bones CRUD app. The little I know about things like AI/LLM is you feed it text so it can learn. If the input is not good then the output is not going to be good either, doesn't matter what it does with the input.
We (as an industry) can't get feature requirements or business logic documented to be interpreted consistently by humans, who understand those problem domains in high fidelity, let alone some computers reading that text. If the translation of requirements to code isn't great, code to LLM to produce new code isn't going to cut it either.
Our industry jokes about all we do is CRUD apps but once an app is mature and beyond simple models, has integrations from a dozen APIs, has customers integrating via APIs, does reporting, needs to guarantee-ish transactions and most importantly is using derived sets of data for billing/invoicing it is much much more than a "CRUD app".
> We (as an industry) can't get feature requirements or business logic documented to be interpreted consistently by humans, who understand those problem domains in high fidelity, let alone some computers reading that text. I
It would be pretty cool if LLM was able to be trained on your company's private code and then you were able to ask it "hey, we're thinking about making XYZ change with one of our vendors/components, what ramifications should we think of" and then it scans all of your code and tells you where stuff is touching it.
I can think of a ton of reasons why were years away from that/small simple breaking things that send the entire thing out the window, but I'd be surprised if we aren't marching towards some version of that next 5-10 years. Think of how many executives in companies will throw money at Microsoft if they promise (they can outright basically lie in the sales pitch and it doesn't even have to work cough Azure cough) it'll scan the entire company's code base and do advanced analytics on it.
It would be pretty cool if LLM was able to be trained on your company's private code and then you were able to ask it "hey, we're thinking about making XYZ change with one of our vendors/components, what ramifications should we think of" and then it scans all of your code and tells you where stuff is touching it.
I can think of a ton of reasons why were years away from that/small simple breaking things that send the entire thing out the window, but I'd be surprised if we aren't marching towards some version of that next 5-10 years. Think of how many executives in companies will throw money at Microsoft if they promise (they can outright basically lie in the sales pitch and it doesn't even have to work cough Azure cough) it'll scan the entire company's code base and do advanced analytics on it.
Lotus had their Approach database which was a wizard Q&A driven system for building basic db apps. For basic stuff it will be fine and will probably enhance the wizard driven stuff out there in basic tools like Excel and Access.
As the requirements become more complex, like having an email sent off after a Create or Update, some SQL stored procedure code needing to be triggered, at this stage because I havent seen enough of the dataset used for training to really know, but they could potentially know enough data to replace a large group of programmers.
What reduces my confidence in that thought, is things like Chap-GPT3 cant even get history right from a wiki page, and cant even code in a specific language.
Its generating pseudo code in some instances, so I think they have a way to go still.
As the requirements become more complex, like having an email sent off after a Create or Update, some SQL stored procedure code needing to be triggered, at this stage because I havent seen enough of the dataset used for training to really know, but they could potentially know enough data to replace a large group of programmers.
What reduces my confidence in that thought, is things like Chap-GPT3 cant even get history right from a wiki page, and cant even code in a specific language.
Its generating pseudo code in some instances, so I think they have a way to go still.
Naah. People have been trying to disrupt CRUD for years already with all these no code tools but nothing comes close to building professional grade CRUD using real programming languages. For prototyping or very basic things we already have tools like bubble. We don't need AI
Every other day I get a new customer opening up with "Hey, I'd like you to build this super simple CRUD app for us, it's basically just a... <insert specs here>". And months later we're still publishing new code & updates because it's never "just this" and never "super simple". So I guess it's going to take a while until AI gets to this point.
Oh, and the other very popular request that we get is "Hey, we have an app built with <insert no-code platform here> and we need to re-build it with traditional ways because we can't get new features in it because the no-code platform doesn't support them..."
Oh, and the other very popular request that we get is "Hey, we have an app built with <insert no-code platform here> and we need to re-build it with traditional ways because we can't get new features in it because the no-code platform doesn't support them..."
IMHO, if no breaking through, we'll probably never get there, based on whatever we have right now in so called AI field.
No impact. CRUD isn't that simple or it would have been automated out a long time ago, AI or no AI. Even low code or no code solutions have not been sufficient to disrupt this industry. There is often business requirements that must be modeled. Not to mention that not all aspects of these types of jobs are 100% focused on CRUD applications but sometimes small utility programs that move documents and data from one place to another. AI isn't at the point where they can learn to investigate how the business works, then ensure that the processes are unaffected as these processes are improved upon.
Similar to the way it currently works with AI generated video or even code. Yes you can create simple stuff just like on retool, but want it to look polished like or be extremely light weight and performant? Still going to have to go in and know enough to fine tune or create custom new components, backends, data models etc.
So basically a parking app that save you a spot, aka low hanging fruit now, sure the ai stuff will eat that market, but want something like Carrot Weather, ToDoist, etc you are going to need someone to build it.
So basically a parking app that save you a spot, aka low hanging fruit now, sure the ai stuff will eat that market, but want something like Carrot Weather, ToDoist, etc you are going to need someone to build it.
Perhaps we'd be able to promptly enter data without predefined structure. AI will help suggest entry form structure progressively. Searching and reporting will be in natural language.
The source code output from ChatGPT looks right but that shouldn't be confused with is right.
Maybe the next generation of text AI models will do it :p
Maybe the next generation of text AI models will do it :p
A team at work is maintaining a 30+ years “CRUD” application. The amount of complicated carefully maintained business logic the application contains is breathtaking. Yes it is a “CRUD” app but it would take many many years to recreate it. And look at SAP. It is also a “CRUD” application. But there is nothing simple about SAP.
In the short term, I don't think current-generation LLMs are particularly useful for CRUD apps.
I honestly have no idea what next-generation LLM performance will look like, but it could _possibly_ be used to generate entire simple CRUD apps.
In the long term this is a bit like asking how ICBMs will affect fortress wall construction.
I honestly have no idea what next-generation LLM performance will look like, but it could _possibly_ be used to generate entire simple CRUD apps.
In the long term this is a bit like asking how ICBMs will affect fortress wall construction.
CRUD only, wow, so easy!!! I'm pretty sure someone has seen the following joke:
The husband told by the wife: "Buy a watermelon, if you see eggs, buy a dozen" before going out. So he went home with a dozen watermelons.
The husband told by the wife: "Buy a watermelon, if you see eggs, buy a dozen" before going out. So he went home with a dozen watermelons.
Without the need for humans, it’ll just be “CUD”
It will be a commodity. Retool is already doing this anyways
What role will AI play in low-code tools like Retool?
I think AI will speed up my building of them.
So much of the world of software development is building variations of custom CRUD applications that take user input, store it and then present it back to the user in various ways allowing them to read, update or delete it.
On top of this is often a layer of other features such as workflow management, notifications etc.
How do you believe this type of software development will be impacted by the advancements in AI in general and LLMs in particular?
Cheers!