Ask HN: How will AI impact the future of CRUD apps?

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.

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".

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.

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?