If someone said “do you want to interview my agents first”, I think the interview would be over. Well maybe the first time I’d be tempted to see if I could tokenmaxx someone else’s agent. After that it would be too much work for the enjiyment. Maybe I should automate that workflow… Oh no they got me!
“Your agents!?! You mean the proprietary models you did not develop, own, or control, and just pay a foreign SaaS company to utilize, built by people much smarter than you, using software coded by people much more competent than you, and can be replaced or removed at any time? Why do I need you again?”
“I like money. Me want money. Give money now. Ooga booga”
The caricature you sketch definitely does exist
But so does the IT hobbyist using an open source model with local hardware and customised training.
Of course i just realise that person wouldn’t call it an agent, they wouldn’t tell you to interview the agent and probably wisely frame it as:
“I develop my own tools and automation systems to free up time and resources for more important projects” at most refer to “machine learning” and absolute avoid using the term “ai”.
It’s called satire
Fair!
If you’re smart, you’d use your personal computer running a local AI model. It’s your expertise that made working prompts because AI on its own is trash. So when you leave, your knowledge goes with you.
Yeah but you’re forgetting one thing that years in IT have taught me. Most people aren’t smart. Most people are dumber than shit, are barely literate, and refuse to learn. They only know “I press this specific button located at this one place for things to happen”. If the button is changed in any way such as a slightly different color or moved by an inch, they are completely lost and helpless
Most people are dumber than shit, are barely literate, and refuse to learn
I’m glad to know I’m not the only one who feels this way. Thank you. I feel slightly more normal today.
How’s the joke by George Carlin go again?
Something like “imagine the intelligence of the average person and then realize that half of people are dumber than that.”
Ugh, you are giving me flashbacks to when the start button stopped saying “start” on it. I lost so much faith in humanity during that upgrade.
Even though programming and IT is a great hobby of mine, I ultimately decided not to pursue it on a university level as I didn’t want to potentially turn my hobby into something I despise. I’ve been regretting this decision a lot throughout my life, but as AI becomes more and more incorporated into the work life of a programmer, I’m starting to feel relief. I would not want to work in that field today.
Programmer here: i actively despise where this field is heading and where it is at right now
i’m studying software engineering rn and honestly i’m not sure i even want to finish my bachelor’s degree. finding an internship (required for the degree) is almost impossible now and even if i do manage to get a job, do i want to spend the rest of my life prompting claude? fuck no!
I am a software dev trained 3 years worked 4. Even i cant find a job. Make of that what you will
I was considering it like 10 years ago when you couldn’t go anywhere on the internet without someone telling you “Programmers make a bajillion dollars pr day, even the bad ones! You should become a programmer!!!”, but I figured it was just companies wanting more applicants to force down the pay.
I think you made the right choice. The IT sector has been undergoing progressive enshittification for decades and it’s really gotten bad with AI.
As someone who made the opposite decision, I get it. And you’re not wrong. I still love what I do when it’s just me and some trusted colleagues, but I’ve had to watch people I trusted deeply in this field fall into just the worst AI holes. Thankfully AI is bad enough that, idk how to put it. I’ve developed an understanding with management and clients that I might not have an email out saying my project is done as fast as the people that do use AI, but my project will actually be done faster. If a solution takes me 2 months to finish, it’ll take a week in testing before hitting production. I’ve seen people declare a solution made in weeks only to spend actual months in testing, having to fix bug after bug. And they’re not even the worst offenders. I know people that will deadass say “looks good enough” and have broken test servers with just actual garbage.
Regret is a funny thing since none of us are fortune tellers. When you realize that every decision you make works the same way, you get a lot more chill about life.
It’s been bad even before that.
But I’m just an old person yelling at clouds.
So I work for a state government and I can code a wee little bit. I found some ways to automate a good chunk of my work with app scripts. Full disclosure, I review them but I couldn’t do it without Gemini. 😅 (That being said I would never trust an agent as an automation for anything)
I made the mistake of telling a coworker or two about the first automation I did. One of them told the boss thinking I’d get kudos.😔 I did and more work.
Haven’t shared any of the automations since. I also built all the ones since to slowly work over the work day so it looks like I’m doing them manually in real time. I also automated all of the data imports using my work email so if I get fired. My account is terminated too and they will all stop functioning.
I gotta say it’s actually made my work day much lighter. Hopefully it won’t catch up with me negatively
Is that khan from khan academy;
An average 22 year old does not have the experience to automate 10 workflows, let alone 100. They also don’t know all the gotchas. This is dumb advice.
Also “automating a workflow” IMO shouldn’t involve having an agent at the end. If tokens and cloud models with selling a dollar for a quarter pricing levels are required for your automated workflow to work, you’ve set up a ticking time bomb.
Use this subsidized garage to churn out regular automation that you can maintain after the merry go round inevitably stops. I was told as much by Microsoft AI specialists themselves. Do not toss out working ci/cd processes in favor of putting a chatbot in the middle, because who knows if you’ll still be able to run that chatbot in a couple of years.
Years? Many models are gone in a few months. If you build 100 automations based on them, you need to upgrade and retest them constantly.
And most business processes are totally fine with a “If this, then that” logic. Companies work like that. The law kind of works like that. No need for expensive and problematic chatbots. Just automate based on simple if loops.
If people were looking at this intelligently, they’d clearly see that AI is heavily subsidized and you can use the subsidized help to fix your broken shit without becoming completely dependent upon it being subsidized forever.
But stupid is as stupid does, so they’ll all just continue to add unhelpful chatbots to their software and token consuming nonsense workflows instead of fixing any quantity of their huge amount of open problems.
Yeah - those people investing trillions in AI companies and datacenters want to earn their trillions back with interest.
Pretty much the same time bomb around any proprietary software.
These fucking morons have convinced themselves that their glorious chat bots are actually able to replace actual work.
They can replace some work, especially debugging. There are a bunch of bug fixes I wouldn’t have made because debugging them was too inane for too little reward. But Claude has no problem writing 14 throw-away scripts to try and reproduce the issue until it succeeds.
Honestly, this. This is exactly the usecase that I’ve found it best for Claude. That and automating a test that I would otherwise have done manually.
The niche I work in (FPGA engineering), the builds take a relatively long time (right now, at my current company my builds are generally 30-60m. Other places I’ve had 6+ hour builds…). A few weeks ago I was testing a corner case I found and was doing it manually, which was tedious. While building a new image that should fix it, I told Claude what I wanted to test and the clanker spit out a script that would have taken me longer to build than the FPGA image would have taken to finish. This way, I also have a somewhat decent (definitely not perfect) script that can take care of testing that corner case pretty well automatically.
That kind of thing is probably the best use case I’ve found for clankers like that.
Translation: “I have no real understanding of anything beyond feeding it into ChatGPT. Please pay me millions of dollars to be a net loss to your organization.”
Laaaaame
First off, they’re not his agents unless he’s got his own LLM running on his own server; and you can’t just show up at a new job with a server all like, yeah, plug this into your network and grant me permissions, it’ll be fine.
Secondly, you can automate workflows without LLMs. Crontab, for instance. Trillions of people have been automating workflows for tens of thousands of years. Agentic automation is more expensive and less reliable than cron.
If someone is showing up at my office and telling me they’re going to use Claude to rename a .PDF I’m going to laugh and then ask them to leave.
You want the future? Don’t be the guy selling shovels during a gold rush. Be the shovel-maker, and then hoard your shovels to inflate demand. Fill your house with shovels. Use shovels instead of plates and cutlery and clothing. Make sure everyone knows that you’re The Shovel King (or Queen) and don’t tell anyone you’re unemployed, no matter how many job interviews you fuck up by talking about shovels.
Why would you go through the hassle of setting Cron when you can ask an LLM running on a server in another continent every minute: “now it’s [time]. If it’s [time_required], then do this”
unless he’s got his own LLM running on his own server
You forgot the part where the LLM also needs to be trained exclusively on the creations of that very person themselves. Everything else is stolen credits.
Calm down DeBeers
Seriously though, I don’t touch llms and just hoard a folder full of various bat and ps files for different things I’ve needed to do repeatedly. And a few txt files of snippets and one liners for less routine items.
I do user facing support more than server wrangling. It works better for my workflow since I don’t know what machine I’ll be on any given minute
Even if the LLM worked flawlessly, you’re just proving your lack of purpose to the company which means they’ll fire you. What do these people even think?!
If you’re automating workflows, you don’t get to take your current employers code with you when you leave.
Are you supposed to just be writing random automations for things that might exist somewhere? Why would anyone think that’s impressive?Either you’re showing me something useless you did for no reason, or you’re asking if I want to see some stolen IP. Neither of these is giving me a particularly good impression.
Much more impressive to have genuine, real contributions to open aource projects… Not those garbage “I fixed 10000 typos in readmes which were actually just word choices and 250 of them got pushed so pls hire me”
Those agents aren’t yours in multiple ways. That’s not how things work.
It’s funny too because as soon as the model changes the behavior shifts. So it’s even less of a flex.
I hope the hiring manager responds with leetcode questions on a whiteboard.
Ah yes, automation, the field that is well known for systemizing processes into chains of predictable machine I/O — the perfect place to inject the mercurial hallucination service I rent.
Aww, poor little guy can’t automate things without agents.
Right? Automation is a fantastic skillset, and when a tool breaks? It takes me a fraction of the time to troubleshoot one I wrote myself than it would some vibe coded bullshit.
And the next tool I write comes out faster and accounts for the edge cases that broke the last one as I understand the system I’m interfacing with better.
“What’s that boss man? You need me to fill out a google sheet twice a day with data from our support ticketing system? No problem!”
Boss man, two weeks after I wrote a script and added it to cron: “Wow, you’re really dependable!”
