AI will replace programmers

The client wants to drag and drop their own personalized excel file with no guaranteed formatting or column order or data contract in order to import their data into our system <3
- Needs more AI to randomly guess what the columns might be
- I love how this is a universal experience.
- Jesus, this gave me war flashbacks.
  
  Please, do elaborate. Let others feast on your suffering.
- Strangely enough we actually solved this problem with AI a few months back. We upload the excel file to Gemini and have a prompt to extract the data we need in a specific json format. And it works surprisingly well.
  
  How well? Bet your life on it well, or "fewer hallucinations than we would have guessed" well? I've considered and toyed around with openAI models for logging supply room check offs in a JSON format and it went better than I hoped but worse than I needed.
  
  It would still have to be in at least somewhat of a consistent format. Even a human would require that.
  If they're just going to write the details however they feel on any particular day and then just expect someone or something to be able to interpret that they're going to have a bad time.
- Do we have the same client?
  
  Everyone has and is that client.
- Or headers. Just unlabeled data in a CSV.
  
  Yep, RIP

I’ve been a professional software developer for over two decades. There is zero chance my job will get taken by an AI any time soon. Anyone who thinks my job is to write code doesn’t understand my job. That’s like saying a bus driver’s job is to turn a steering wheel.
My job is to turn vague ideas and nondescript feelings into APIs and (sometimes) UIs, then turn those into specs, then split those into tasks, then sometimes I’ll write the code for them and sometimes someone else does. About 90% of my time is turning ideas into plans, and about 10% of my time is turning those plans into code.
When I was young and was a junior engineer, my job was more to receive the specs from the senior engineers and turn that into code, but even then, I was still designing my own stuff. Maybe more like 40/60 time instead of 90/10.
Now that I’m a grizzled old man forged in the fires of task management software, I’m doing almost all of the design work myself. I manage a project that has about 250,000 lines of code. An AI isn’t going to be able to build new features into that, let alone decide which features to build in the first place.
- Heh, that won't stop a C-level from thinking that you just write code.
  
  Yeah, that’s probably true. Remember how all the execs decided to replace cashiers with robots, then the stores started losing money because a. it made stealing a lot easier and b. people would avoid stores that only had self-checkout robots and never had anyone to help you because a robot doesn’t know where the flour is. Now the self checkouts are being decommissioned and we’re going back to regular human cashiers. It turns out cashiers do more than just scan barcodes. But, upper management didn’t get to where they are by being smart.
  
  That's why the best places to work tend to be the places where your CEO has had your job before.
- That’s like saying a bus driver’s job is to turn a steering wheel.
  That's a good analogy, I will use that.
  
  Full self driving.... NOW.
  See. It can turn the steering wheel on is own. Feature complete!!
  
  The problem is that the CEO who believes he just writes code probably believes all a bus driver does is turn the steering wheel.
- It could also be possible that AI doesn't write code but becomes the new software. Hook it up to libraries, contractions, databases and pump it full of the verbatim ramblings from the client, sales guy and manager. Sure, it costs a magnitude more power to run but it's up in no time.
  Then hire a few consultants that used to be senior engineers for an outrageous amount to trace the weird hallucinations and replace the mission critical part with real code.
  The future is going to be wacko.
  
  That's the way it's going now. People are making agentic stuff. Even in my company people are experimenting on something like that. Setup an agent, hook it up to the APIs and let loose.
- It won't automate your job, but isn't it already replacing junior devs via team downsizing?
  
  Any place that is replacing junior devs with AI is probably going to really regret it when they have no senior devs in a few years. Being a junior dev in a team is kind of like an apprenticeship. You learn the trade, but you also learn the shop. Then when the senior dev moves on, you have all that knowledge and can step into the role of senior dev. If a team decides to not have junior devs anymore, then they’ll have no one to take over when a senior dev leaves.
  So the answer is yes, it is already replacing junior devs, but that’s only because management hasn’t learned how bad of an idea that is yet. Ultimately, it will cost them more through losing foundational team knowledge.
  You also have to hold an AI’s hand the entire way through coding something, whereas you can kind of just let a junior dev go do their own thing, and eventually they’ll probably get it right. An AI “agent” tries to hold its own hand, but that doesn’t seem to work out usually when I’ve tried it. It starts making changes that are really bad, then just seems to always double down and eventually make a huge mess.
- Could AI allow you to write code in python, and then turn the python into a static language with static variables at least?
  
  A language to do what?
  That is just programming with extra steps
  
  But the main benefits of static typing is in making the programming part easier. What do you gain from translating dynamically typed languages into a statically typed language?
  
  I don't see how that's going to work out well. That's asking to end up with a mess that you're just going to have to rewrite anyways.
  I do not even have a complete hatred for AI like a lot of folks do, but I don't trust it that much (nor should anyone).
  You'd be better off with an actual deterministic transpiler for that (think TypeScript -> JS but the other way around I suppose), not something with a ton of random variables like an AI.

No, the customer wants a button that does a very specific thing.
He can't tell you what that is, though. You're the expert!
Also, can you put in more ads? And make it so the users can't close the tab until they bought something.
- You're the expert!
  I can do absolutely anything. I'm an expert!
- make it so the users can't close the tab until they bought something.
  Why do they always request that? They've never seen any website on the internet work like that yet they actually think that they are the first people to come up with that idea.

This is when the AI, in a microsecond, decided to destroy the human race.
- Not gonna lie, I don't really blame the AI.
  
  This is why we definitely shouldn’t rewrite the nuclear launch software. A project manager could unintentionally push a programmer into justifiably ending the fucking planet.
- That's when you load all middle managers onto a spaceship and send them off under the guise of colonising a new world.

- If those project managers could read, they would actually be able to use Jira.
  I regret what I wrote.
  I realised, just being able to read, doesn't make sure they can actually use Jira

Another crap meme made with AI, yay, I love AI slop

I had a client once explain to me that his request for the 75% redesign of his mobile app would be simple because "it's just 3 pages"
That was the exact quote
I know that was hardly related to the post, but it reminded me of that and I needed to vent to my therapist (aka strangers on Lemmy)
- I feel you. Just ended coding "a little special case" that resulted in dozens of files changed, all because I refused to make it with dirty hidden hack, and that was a clear-cut technical if-branching even, no vague ideas
  Talking to a client is times that amount of hurdle

Button that does something? That’s too advanced for me, I’ll use a library
- Help, Debian has libbutton only in 1.4.3 and libdosomething is not in my repo. I compiled libdosomething from source, but now it needs libbutton >= 2.4.1 and compiling that version of libbutton fails, as my GCC and make are too old and incompatible!
  I already tried it on my other PC, but that isn't based on glibc, which makes all these dependencies even worse...
  
  Have you tried unplugging your computer, going out into the woods, and returning to monke?
- cries in left_pad
  It's kind of astonishing how many people leaned on that library just to add fucking spaces to strings
  
  It should be in the standard library anyway. Why the hell is it not?!
  I mean yeah, I can write my own function to do the same thing and probably I've done it at some point in some coding exercise as a beginner, but this seems like such a common thing to use, it should be in the standard library of any sane language.

You know what we, in the industry, call a detailed specification fo requirements detailed enough to produce software? Code.

Surprised there's no one in the comments going bat shit crazy that this was made by AI. Are we not doing that anymore?
- You just had to wait 2 more hours for that.
- Well, the table sucks. I can't draw and I could do better.
- Character development ❤️

Managers about to find out the hard way that all the requirements are in the brains of those they laid off.
I’m sure coding bootcamp and AI will turn them into leet hax0rs.
- Definitely happening at at least one major company I'm familiar with.
  Requirements and everything else.
- I feel this in my bones. Even before the recent round of restructuring we've had a significant about of turnover. Our infrastructure is a massive rube golberg machine with multiple houses of cards built on top of it. Institutional knowledge was never written down and it has been leaving the company at an accelerating rate over the past 5 years. Tons of "new blood" making lots of assumptions on how things work is resulting in... humorous end results.

presses button; nothing happens
"Well see here! I wanted that button to do something!"
"Oh but it did! It wasted your time as well as mine!"
- "But I want it to do something other than that!"

AI slop image, for this gag?
- How queer, I must report to my supervisor posthaste!

AI:

Whoa whoa, hold on there! You can't expect a product manager to come up with such detailed specs!
- I am a product manager that loves coming up with detailed specs. How else will I actually get what I want? If you care about some specific behavior/outcome you must specify it. This logic is lost on my leadership.

Customer requirements are basically always “I want what my Excel sheet used to do”.
- "You mean you want it to corrupt your data and end up with conflicting changes once you share it?"
- I want faster horses.
- Make sure to convert 1/2 to February
- TBF if the spreadsheet works, why change anything?
  
  That depends. Is the spreadsheet doing what it's intended for, or one of the hundreds of things it can do but really shouldn't?
  I've made my fair share of spreadsheets, including time tracking, vehicle scheduling, email automations, map integrations, god forgive me even 'databases' - all of that because no one was willing to pay for a proper solution.
  So I'm not saying spreadsheets can't or shouldn't ever be used for those things, but a dedicated solution for a problem may indeed be a good reason to change things.
  
  <.<
  We have an excel. It’s more… a group of excels? They call each other. Opening and running VBA within each other. The excels are “run” nightly, and take well over an hour to do their thing.
  Should we just keep it in Excel?
  
  They change it because they need more money. (talking about Excel here)
  And soon you'll see ads in the fn suggestions drop down
  and ads in unused cells
  and ads while the Solver is running (made in VisualBasic, running on top of WebAssembly, running on interpreted Python, using a Java Interpreter on your Web Browser to make sure you watch the ad for as long as possible)
  And that's why we need to make our own spreadsheet software.

AI Project Manager: Create a button on a webpage that, when clicked, displays an alert saying "Hello World!"
AI Programmer: "What a sensible requirement! Here you go."
AI Billing Department: "Project completed, that'll be 10 million dollars."
Client AI Payments Department: "Sounds right, paid!"
- AI Quarterly Call Bot: Delivery is on time and synergy is high!
  AI investment selector: This company looks profitable. Purchase!

"Can we jazz it up a bit?"
This is a real request from a real manager. They have played us for absolute fools.
- My favorite was to make it “crisp”. My boss was sympathetic when I joked about it being “crispy and delicious” but he still acted like we basically should have known what that means.

"Y'know, I've been thinking... The app is missing a couple of things, like This, and That, and it should also do This after That, but not That after This, and maybe even navigate to The Other Thing after 3 Launch events, while also not doing that if the user is under a Pisces moon in the 4th Year of Wilting..."
"So... you want a Rate the App pop-up with specific trigger conditions?"
"What?! No! I want one of those prompts with the stars and the redirect to the Store which lets people post reviews of the app, what are you even talking about?!"
AI Junior Dev: short-circuits
- It wouldn't short circuit, it would just say "OK, no problem!" and then output bullshit.

Me: I need spec -- not just trust code Manager: You always make unnecessary demands, I'm replacing you AI: I would be happy to help you, if you could provide spec? Manager: god fuckin dammit
I honestly sometimes think to go into business myself just so I can write contracts that say "you will give us a fucking spec" and just keep billing while they fuck around not providing a spec

My wife had her first meeting with Chat Gpt today.
She went from a random question about her job to the AI offering to taking care of her LinkedIn page, and promoting alternative positions for her.
It feels to me the product manager is in trouble too.

I'm sure AI can solve the problem!
- Current LLMs would end that sketch soon, agreeing with everything the client wants. Granted, they wouldn't be able to produce, but as far as the expert narrowing down the issues of the request, ChatGPT would be all excited about making it happen.
  The hardest thing to do with an LLM is to get it to disagree with you. Even with a system prompt. The training deep down to make the user happy with the results is too embedded to undo.
  
  The hardest thing to do with an LLM is to get it to disagree with you.
  Yeah, I occasionally use conversational AI and its really hard to let the AI have any agency in the story because they usually just go ahead with whatever you write

A highly respected school teacher of mine was known to say, "Say what you mean; mean what you say."
- This is really one of the best advice about communication ever.

Stupid PM. Forgot to say the button needs to be corn-flower blue.
- Make pop.

Great! Software isn't bloated, convoluted, incomprehensible, fragile enough already!

Yeah, I have a guess that it didn't even fullly understood the prompt behind this slop either...

Just wait until QA gets a hold of it

The DWIM button.
Coming soon after the Neuralink implant.

AI can't replace programmers right now, but I've said all through my software dev career that our ultimate goal is to eliminate our jobs. Software will eventually be able to understand human language and think of all the right questions to ask to turn "Customer wants a button that does something" into an actual spec that generates fully usable code. It's just a matter of time. Mocking AI based on what it currently can't do is like mocking airplanes because of what they couldn't do in the 1920s.
- I had a number of points to discuss, but they pale before this:
  Software will eventually be able to understand human language
  First, someone surely must have tried to code it, but I never heard of any system like that. Second and more important: anyone understands how we understand? And how the distance between understanding and communicating is covered? Someone? Anyone?
  And before some smart person tries for the thousand's time this "but computers will get bettah" shit of argument: even with the whole task of putting it to code aside, we know shit about how we think, understand and speak, that's coming from me having Master's degree in linguistics
  
  Yes, the main problem with developing AI is that we really don't understand how we think. Current AI doesn't understand anything, it just imitates human output by processing a vast amount of existing output. But we do know a lot more now about how we think, understand and speak than we did a hundred years ago, and as a linguist you know this work isn't standing still,. Compare it with genetics - 70 years ago we didn't even know about DNA, and now we can splice genes. The fact that there's still a lot of baseline work to do shouldn't cast doubt on the goal, should it?
  
  I just want to point out that this whole AI thing started with people not understanding how it works. I get your point although I think it's stumbling into progress rather than understanding that will be a method until we do.
- Or like mocking Moon colonization in 1970

How about... AI replaces government officials! A lot cheaper. Might actually get things right. And how could it fuck up any worse than what we have?
- A lot. The answer is it could fuck things up a lot worse than what we have.
- "Goverment bad durr durr"

Good, I hope so. Whatever puts an end to the hipster, activist dev is a solid win.