[superultra]

Had Waffle House with some friends who mostly work in blue collar industries. One guy who works at a timber mill used Claude code to redo their ordering system. Took him about a month to go from knowing nothing about Claude Code to finishing the system. Basically just copied a proprietary software product that costs them upward $20k a year. They’re keeping that other product to cross check but so far the Claude coded item works great, and is of course more custom to their business. The dudes a hero at work because the system is heads and tails better.

Obviously caveat emperor but there are a lot of real world scenarios like this.

I think Anthropic and OpenAi are trying to all cool and apple-y with their branding but these use cases are just tools getting work done. Most normal people don’t need or want AGI, or even AI slop videos. They just want their invoicing system to just f-ing work for a change.

[everforward]

> They just want their invoicing system to just f-ing work for a change.

Time will tell, but I'm dubious this will hold longer-term. I don't doubt that Claude can write the code, but I am dubious Claude can manage it sanely. Does it have backups? Does the guy that wrote it know how to restore those, or can Claude do it? Can Claude upgrade the backend and/or migrate the data when the backend changes, or is this going to be running known CVEs in a month?

This has sort of always been a thing via hiring CS students as interns. I don't doubt most of them could jam out something that looks like Slack or Gmail. The problems aren't apparent immediately, they become apparent when you realize it doesn't handle invalid responses well and the backups are hosed so you just lost a bunch of data.

[pjc50]

I'm converging on this as the real end state: it's a "better Excel" for general business work. And has some of the same limitations - maintainability and security. But there are also plenty of small businesses that run off a shared Excel spreadsheet and a few mailboxes.

Nobody ever really solved making CRUD apps easier through better frameworks. So now we have a tool to spit out framework gunk, and suddenly everyone can have their own app.

[elevation]

> caveat emperor

s/emperor/emptor

I hope your friend's company spends $20K to harden the deployment of the new app so it doesn't become a deep liability.

[windexh8er]

Keep dreaming!

The best part is is that they'll get popped because of it and have zero clue. Anyone building in any frontier provider currently, but has little background in software, is creating all kinds of new liabilities that didn't exist before.

In a school district where I live the IT department developed a password distribution app using Gemini on Google App Script (they didn't even need this part), sent out links with B64 encoded JSON that included: student name, student email, parent email and student password. Yet, when I found it and told them all the ways that it was technically a breach in our state they ran to their 2-bit "cyber security experts" and "legal". They were far more concerned with CYA than understanding the hole they dug themselves. And all of the advice they got back was that it wasn't a breach. They claimed their DPA with Google protected them. I explained how email works and they just ignored me, likely because in our state they are bound by GDPA and won't ever engage in a legitimate conversation via email.

The kicker here is they pay for an IDP with built-in mechanisms for password resets (that was the reason for building this: to reset students passwords). One of their cyber security "experts" (a lone guy who has zero credentials from what I found) told them that password resets using the IDP was "not recommended". When pressed on that they were, again, silent.

LLMs are creating a huge mess for people now empowered to go well beyond their capabilities and understanding. It's a second coming of the golden age of shitty software that's riddled with even the most basic of security flaws.

[seanw444]

I'm just going to keep building software mostly traditionally, while using "AI" to help me research things quicker (might as well use it while it's here), survive the shitpocalypse, and then laugh as traditional-minded developers become a scarce sought-after resource again.

Either way, the instability of this industry due to the insane amounts of cargo culting every time <insert big thing> comes along has made me really question whether I want to stick around.

[heavyset_go]

> Either way, the instability of this industry due to the insane amounts of cargo culting every time <insert big thing> comes along has made me really question whether I want to stick around.

Whatever you do, don't click this link: https://github.com/garrytan/gstack/

[superultra]

I think this is where a lot of freelance contractors could pivot to - basically "last mile" coding, where the LLM does the front end work, and then high hourly pay engineers come in and fix the work. it'd still be cheaper than a lot of the industry niche software that is usually pretty bad.

[superultra]

thanks for the correction

I hear you but at least as my bud described it, the software that most of the timber mill industry uses is buggy as hell, crashes all the time, and makes mistakes. One would wonder if even the licensed software is hardened.