Microsoft's hiring shift: Fewer generalists, more AI-driven roles

[-]

Full-Spectral@reddit

I feel like I'm taking crazy pills reading this section lately. We aren't going to have to wait for AGI to become aware and kill us all. We'll all be sitting around prompting LLM's to tell us what prompt to prompt what other LLM to tell us whether the previous LLM answer was the right prompt for the other LLM.

[-]

Brilliant_Lobster213@reddit

AGI was never a danger to society. The real danger is capitalists getting rid of the working class

[-]

Full-Spectral@reddit

Come on, that's kind of a silly argument. Those people know where the real money is.

Some of them may be greedy and completely without conscience, but they aren't stupid. They know perfectly well that a broad working class is in their best economic interests, and moving more people upwards into the various strata of the working class even more so.

Poor people don't buy lots of stuff, and people buying lots of stuff is how most rich people get rich. And our being middle class is no real threat to them.

You could argue that some of them really like keeping middle class people distracted by shiny things and looking away from the real issues, and I wouldn't argue with that. But that's not the same thing at all. And I'd also argue that a lot of people interested in (and contributing to) that aren't captains of industry either.

[-]

blamelessfriend@reddit

Come on, that's kind of a silly argument. Those people know where the real money is.

brother the billionaires are building impenetrable bunkers instead of renewable energy.

how anyone still has confidence the richest among us are the smartest is beyond me.

[-]

Full-Spectral@reddit

They don't have to be the smartest, they hire people to do that. That's one of the big benefits of being rich. Though some of them probably are quite good at what they do and quite intelligent.

It's easy to be cynical and edgy about it all. But plenty of the super-rich are very much into renewable energy, and plenty are on the other side. They aren't any more monolithic than we are. And the vast majority I'm pretty sure don't live in impenetrable bunkers.

[-]

superxpro12@reddit

In today's world, money equals speech equals power. With the current shift the entire world seems to be taking to the right, i simply dont see the benevolent billionaires doing the right thing. They will only do what pads their power, which means, in most cases (and see the trump inauguration for the whiplash all the US billionaires did to align behind the current admin) aligning with authoritarianism.

[-]

Dean_Roddey@reddit

Did Trump magical wish himself into office, or did a bunch of people vote for him? We still control who gets into office. If we choose to elect people who aren't going to work in our interests, whose fault is that?

Ultimately, we still control who gets to be in office, and can get rid of ones that are in office if we choose. That's our job in a democracy. If we don't, whose fault is that? Make schoozing with bazzilionaires a political third rail and it'll stop.

None of this has to depend on either billionaires or politicians wanting or not wanting to do the right thing. Politicians want to stay in office, and that's all it takes. And no amount of political funding could get someone into the White House if we as a people decided we don't them to be.

[-]

superxpro12@reddit

I genuinely still have trouble understanding how he carried every single swing state. Every one. Every single one of his rallies were half-empty or worse.

And with all the circumstantial evidence from him and musk bragging about fucking with the voting machines.... it's not logical.

[-]

EveryQuantityEver@reddit

Combine a global anti incumbent sentiment with a bunch of racism (seriously, one cannot discount the amount that’s the Trump campaign was racist)

[-]

EveryQuantityEver@reddit

Some of them may be greedy and completely without conscience, but they aren't stupid. They know perfectly well that a broad working class is in their best economic interests

Literally every action taken by these people and the companies they own disproves this

[-]

bring_back_the_v10s@reddit

We'll all be sitting around prompting LLM's to tell us what prompt to prompt what other LLM to tell us whether the previous LLM answer was the right prompt for the other LLM.

Heh my life summed up right now.

[-]

therippa@reddit

I was thinking this when I saw the other day that pewdiepie runs a pretty insane local llm setup (10x4090s) and has a "committee" of 67 agent personalities that work together to answer his questions. That seems like so much work to not just doing the thinking yourself.

[-]

HappyAngrySquid@reddit

Me no want think tho. Hurt brain.

[-]

syklemil@reddit

We'll all be sitting around prompting LLM's to tell us what prompt to prompt what other LLM to tell us whether the previous LLM answer was the right prompt for the other LLM.

That's some serious rat in a cage with a slot machine energy. I guess we're kinda primed by our pocket dopamine dispensers as well.

[-]

MadKian@reddit

You joke, but I did a GCP Model Garden training and the last exercise literally made me use a model that compares the answers of two other models and gives them a comparison score.

Why the fuck would I trust or need an LLM for that?

[-]

boobsbr@reddit

It's LLMs all the way down.

[-]

Miserable_Ad7246@reddit

Imho generalist get much more value from AI. You roughly know what needs to be done, and AI helps you figure out the details. Specialist on the other hand knows most of it anyways. If anything generalist is more flexible and has a headstart on many more problems.

[-]

graph-crawler@reddit

You assume there's no need to double check AI output...

Also a specialist will shine on novel things, that LLM can't do

[-]

Miserable_Ad7246@reddit

You assume I assume.

I'm doing some hard software engineering, some of that is very new to me, and LLM helps me out to develop solutions from first principals and fundamental knowledge. I broadly know what I want to achieve and roughly what need to be done, but I have little know-how on how to achieve it.

LLM runs few deep researches for me to give me more details about the topic (I also read the sources) I when spend some time to figure out all the details and new unknowns. I when use LLM to generate some code and bounce back ideas. I when test things out and if all the i's have the dots and al the t's are crossed I'm happy to push that to production.

As a generalist I had to learn lots of fundamental knowledge to allow me easily move between stacks and problems. LLM now helps me to figure out the details and solve tactical issues.

[-]

bobsbitchtitz@reddit

As a devops eng/ platform eng as a generalist AI is amazing but also retarded. I use it to help me craft the correct terminology and google searches before I trust anything it says

[-]

Full-Spectral@reddit

Which is why I just do the searches myself. I'm quite a good intelligence, and I can simulate human interactions far better than an LLM.

[-]

diegoeche@reddit

I feel exactly the same. I have seen super smart guys getting blocked (even with access to AI tools) just because they are afraid of getting out of their expertise area.

[-]

red_planet_smasher@reddit

AI is what has let me more easily get out of my area of expertise. Look out devops, here I come! No idea if I’m a generalist or a specialist though, I’ve just been coding for a few decades.

[-]

LeagueOfLegendsAcc@reddit

If you aren't just copy pasting the output or letting an agent just run wild you aren't really vibe coding. Just using an AI assistant. I learned to code over a decade ago personally and it has allowed me to iterate faster than ever before. And my debugging skills have never been better.

[-]

knome@reddit

even more shit, shovelware and bloat going into their offerings, eh?

[-]

Halkcyon@reddit

Every OS release where I get more Copilot advertising spam in my paid, professional product makes me investigate dual-booting some Linux distro as my daily driver going forward.

[-]

HappyAngrySquid@reddit

I switched 5 years ago. I don’t miss a single thing.

[-]

chicknfly@reddit

Honestly, unless you’re playing games that are not on Linux or prohibited in VM’s (e.g., Apex Legends), or you have Os-specific software, then there’s no reason to be on Windows

[-]

Mrzozelow@reddit

I switched a few months ago, now Windows is my backup for playing a few anti cheat games. Everything else can be done in Linux.

[-]

pennyell@reddit

Linux Subsystem for Windows when?

[-]

d3xx3rDE@reddit

This is the way

[-]

Halkcyon@reddit

Yeah, I work on some side projects where Windows is my deployment target or online games that don't support Linux, but otherwise most of my time is spent on games that can run in Proton, Firefox, and vscode, so most of my needs are met.

[-]

Familiar-Level-261@reddit

I especially love the teams "do you want to add copilot icon" with no option to refuse, just delay.

MS have idea of consent of an average rapist

[-]

AM_Dog_IRL@reddit

I switched to Bazzite months ago and it's been an almost flawless experience

[-]

dethswatch@reddit

my favorite, at 3am when I log into their SERVER version to fix something is being assaulted by sports and games in the menus.

[-]

yanitrix@reddit

I switched like a month ago to cachyos. There were some issues/gotchas. but in the end nothing that couldn't be solved with google or chatgpt. Dependeing on the software you use it can be easier or harder but I think it's worth it in the end.

[-]

txdv@reddit

with codex, gemini-cli, claude code it should be really fantastic, because it can literally iterate until it fixes everything and it works

[-]

Flaky-Restaurant-392@reddit

It’s been a year for me on Ubuntu. No regrets.

[-]

blind3rdeye@reddit

yeah, I set up a dual-booting on my home computer last year - but as it turn out, I never once actually wanted or needed to boot into Windows. So after from some testing and checking of things shortly after the initial install, the Windows drive is now just left to rot.

So do it. Hopefully you're lucky enough to pick a distro that suits you first try.

[-]

GraciaEtScientia@reddit

I can tell I need to get to bed based on the fact that I read this as "makes me investigate dual-boofing some linux distro"

[-]

International_Cell_3@reddit

Why aren't you already

[-]

Halkcyon@reddit

Friction.

[-]

Mainmeowmix@reddit

I don't miss windows one bit. Well worth the switch.

[-]

mekkr_@reddit

I only keep windows to play a couple games that require root kits now. All my normal pc stuff is done via mint. If it weren’t for those few games I’d be never using windows again

[-]

jusas@reddit

Good idea. I switched a year ago and never looked back.

[-]

Plank_With_A_Nail_In@reddit

Their profits increased by 25% last year.

[-]

thy_bucket_for_thee@reddit

How else can you justify growth rates if you don't force feed these products to users?

[-]

sisyphus@reddit

On the bright side most MS software can't get much worse anyway.

[-]

pydry@reddit

At this point their share price simply cant be justified unless their AI lets half of their customers lay off a bunch of staff.

So, theyre invested in the lie on multiple levels.

[-]

fire_in_the_theater@reddit

see look, we forced everyone to use AI and then AI number went up!

AGI when???

[-]

bring_back_the_v10s@reddit

Unless science finally understands the fundamental workings of human cognition and how to replicate that in software, whoever says AGI is coming soon is a delusional lunatic who doesn't know what he's talking about. Because cognition is obviously not as simple as a statistical model.

[-]

CurtainDog@reddit

I don't know why you think it's necessary to understand something in order to create it. It's an utterly illogical position that can be countered with the briefest moment of reflection.

Do you think bro understood combustion in order to rub two sticks together?

[-]

bring_back_the_v10s@reddit

You're not seriously putting combustion and human cognition in the same category are you?

[-]

Bakoro@reddit

It's closer than you think.

As it turns out, transformers are mathematically analogous to brain structures, despite not being designed to be brain-like at all.

Transformers, or transformer-like architectures are likely to be a necessary, but not sufficient part of an artificial general intelligence.

As I've been saying for years now, LLMs are not the whole digital brain, but they are most likely the hub that the digital brain will be built around.

As far as understanding human cognition, we do have a pretty good idea about how it works at a mechanical level. Over the past ~5 years, scientists have learned a lot about biological neural structures, and it turns out that biological neurons are more complicated than our historical understanding.

Part of power of the biological brain is taking advantage of chemistry and physics to essentially get nearly free work. Similarly, chemical reactions helps with massive parallelization and alternate processing modes.

Beating the power efficiency of chemical processes for processing is going to be somewhere between difficult and impossible.
To fully represent a biological brain's activity is going to take both algorithmic improvements, and more efficient hardware.

Scientists have already successfully mapped and simulated a nematode brain and a fruit fly brain. They're working on a mouse brain next.

On the hardware side, there are multiple companies that are designing artificial neurons that are much more closely aligned with biological neurons, and at least one company is working on interfacing biological neurons with silicon processing.

The hardware/software sides of progress will converge, and we will have something approximating a biological brain in a matter of a decade or so.

Even then, as I will continually bring up: we don't actually need AGI for radical social changes. We don't need conscious robots for dramatic social changes.
Just a few key domain specific super intelligences, some of which already exist, and a few "good enough" models, some of which already exist, and we can end up with a totally different kind of economy.

[-]

bring_back_the_v10s@reddit

Scientists have already successfully mapped and simulated a nematode brain and a fruit fly brain. They're working on a mouse brain next.

That's not cognition. We're not even close.

No, scientists don't understand cognition. Stop being delusional for your own sake.

[-]

StupidPencil@reddit

Part of power of the biological brain is taking advantage of chemistry and physics to essentially get nearly free work. Similarly, chemical reactions helps with massive parallelization and alternate processing modes.

Can you elaborate more on this? Just some links to wiki pages or research papers are also fine. Just curious what recent breakthrough I missed.

[-]

Bakoro@reddit

Some of the papers I was thinking of area bit older than I remembered because I forgot to account for the lost years, but discusses the efficiency of neural processing, and the chemistry of learning:

https://onlinelibrary.wiley.com/doi/full/10.1002/jnr.24131
https://pmc.ncbi.nlm.nih.gov/articles/PMC4005942/

This one talks about dopamine as adaptive learning rate:
https://www.nature.com/articles/s41586-022-05614-z

There's also this thing which discusses calcium and calcium binding ions. It's primarily concerned with the things that can go wrong, but it's interesting.

https://medicine.yale.edu/news-article/study-reveals-how-calcium-channel-mutations-lead-to-cognitive-dysfunction/

And there's a bunch of stuff about the multiple roles astrocytes play, the chemical support glial cells offer neurons.

I don't have just one paper that explains it all, but when you put it all together, there's a lot of recycling and multiple roles that chemicals seem to play in the brain.

[-]

Designer-Relative-67@reddit

"Transformers are mathematically analogous to brain structures" is total horseshit. Unless youre using an extremely loose definition of analogous. I guess they could be similar in that they process data through connected nodes, which is so general its kinda useless. But its gonna be longer than you think, and will look nothing like an LLM.

[-]

Bakoro@reddit

Okay but you're wrong though.

https://arxiv.org/abs/2112.04035

[-]

NuncioBitis@reddit

AI - when you want garbage out for your garbage in

[-]

CherryLongjump1989@reddit

What the fuck does that even mean? How is an AI-driven role not just a under-qualified generalist?

[-]

pecesiqueira@reddit

AI-driven is cheaper

[-]

wasdninja@reddit

Because it doesn't work.

[-]

Girth@reddit

see, you can't just point out exactly why this is idiotic. this is why you will never make it in corporate management.

[-]

Coroebus@reddit

The perf reviews will be: "Not a team player, lacks entrepreneurial spirit, too attached to status quo. Unsuitable for promotion. Second round layoff target"

[-]

PositiveAnimal4181@reddit

I can't wait for this bubble to pop!

[-]

chucker23n@reddit

That mindset is already visible inside Microsoft. Nadella illustrated this by talking about an executive overseeing fiber networking. In a bid to meet growing demand for cloud computing, she used AI agents to automate DevOps maintenance, scaling operations without having to hire more people.

So this means

a. not a whole lot; some poor soul was forced to ask an LLM what DevOps maintenance looked like, then ignored the result and did it properly, or

b. they're really dumb and letting an LLM randomly determine how production servers at Azure should be maintained

Either way, a fantastic way of saying, "please do not work here".

[-]

the_gnarts@reddit

b. they're really dumb and letting an LLM randomly determine how production servers at Azure should be maintained

After working with Azure for more than a year, I’ll believe that in an instant. Perhaps it’s time to modify Hanlon’s razor:

Never attribute to incompetence that which is adequately explained by overconfident use of an LLM.

[-]

JanB1@reddit

Call it Hinton's Razor or something, I don't know.

[-]

Imatros@reddit

Gnarts Razor

[-]

BroBroMate@reddit

That's "The Gnarts" you peasant.

[-]

ModernRonin@reddit

Never attribute to incompetence that which is adequately explained by overconfident use of an LLM.

They're the same picture.

[-]

sebovzeoueb@reddit

Yeah, I've used Azure and I'm also voting b, although, it was like that before as well

[-]

radarsat1@reddit

I mean this would explain some issues I've experienced (not to mention attempts to open related tickets ..)

[-]

DreamHollow4219@reddit

I honestly can't wait for Microsoft's overconfidence in AI to be their downfall.

The company has been going downhill for years, only a matter of time before companies like Apple really give them a challenge that frightens them.

[-]

SpareIntroduction721@reddit

AI==Overseas hiring

[-]

GraciaEtScientia@reddit

AI: All Indians

[-]

BroBroMate@reddit

I thought it was Actually Indians? Both work.

[-]

GraciaEtScientia@reddit

I asked the AI to settle it, it chose your logic:

"The truth is it's "Actually Indians" - and here's why:

"All Indians" would imply that every single Indian person is somehow involved in AI, which is statistically impossible because that would mean approximately 1.4 billion people are all sitting at computers pretending to be ChatGPT right now. Where would they find the time? Who's running the restaurants?"

[-]

BroBroMate@reddit

Haha, thanks AI. Or maybe I should say... RANJIT!

[-]

BroBroMate@reddit

Kool-aid for everybody!

[-]

fuckthiscode@reddit

Another excellent decision from corporate. There's on such a roll, I compiled a list of software Microsoft introduced in the last 15 years that's considered good:

[-]

BroBroMate@reddit

You missed one. Wait, I mean none.

[-]

gareththegeek@reddit

Gotta get all in on the bubble

[-]

jelly_cake@reddit

Yep. Gonna be a very profitable time to be a virus writer in the next couple years, assuming they're not just bullshitting.

[-]

GraciaEtScientia@reddit

I can't wait for Vibe-Virusses and Vibe-Hacking to become a thing.

[-]

Paradox@reddit

And then we'll all install Norton Anti-Viberus to stop it

[-]

GraciaEtScientia@reddit

Yeah, or McAviberus, whichever you prefer.

[-]

ModernRonin@reddit

"This will end well."

[-]

lKrauzer@reddit

I'm glad I left Windows, constant breakages on recent updates since more than 30% of the update code is being written by AI, so fun.

[-]

MatsSvensson@reddit

^(> Click to view comment)

[-]

Latter-Pen5619@reddit

"More leverage per head" is the real headline. AI isn’t just changing who gets hired. It’s changing how orgs think about headcount entirely.

[-]

GregBahm@reddit

Headline is kind of funny because in 2025, the juniors doing AI are the new generalists. I expect them to embellish it on their resumes, but the kids who style themselves as the lords of AI aren't doing anything more sophisticated than asking ChatGPT what I used to ask Stack Overflow.

Which is fine.

For every 10 resumes I've looked at like that, there are one or two that don't mention AI at all. I don't know what the writers of those resumes are thinking. I'm not expecting some level 60 junior CS grad to have been studying AI in a lab for 12 years. I expect their terrified professors probably tried to put up some rule on the kids demanding they never use AI, and the ones I hired are the ones that were savvy enough to ignore that.

[-]

adreamofhodor@reddit

If you’re at the forefront of developing with AI, it’s definitely more sophisticated than just typing to ChatGPT stack overflow questions.
Not that that’s not a good use case, but agentic AI has legitimately been game changing for the speed of my development.

[-]

GregBahm@reddit

Yeah okay let me just go down to the "sophisticated AI developer store" and pick up a couple dozen sophisticated AI developers on the way home.

The reality is that lots of junior level engineering hires at Microsoft barely even know how to program at all. It's been that way for decades. Schools teach kids how to solve problems that have already been solved before, but the job is to solve problems that have never been solved before, so it's unreasonable to expect an entry level engineer to know how to do their job.

So the goal in hiring is to at least find some kid that expresses a willingness to try learning. That's enough for success in this situation. But sometimes it's still asking too much.

[-]

TitaniumFoil@reddit

But it's not more sophisticated than actually programming...

[-]

edgmnt_net@reddit

Looking at Microsoft's careers site and skimming over a few JDs, at least some don't seem to demand AI skills at all. Maybe you're talking specifically about junior positions, perhaps in a particular subfield?

I'm not working for Microsoft, but so far I've also been relatively unconcerned with AI and I don't see that changing soon for the kind of work that I do.