AI and white collar work
I’m really not sure what the future of white collar work is. I worked hard to break into this industry and it seems the common sentiment is in 18 months all of this will be automated away by some Claude agent. If that’s the case what’s even the point. Might as just study EE or get a trade. I mean 50+ million jobs could potentially evaporate over the coming years and it’s just business as usual. I think I’ve been consuming too much doomer content because I’ve been consistently anxious over what lies ahead for this industry and any other office job. Any thoughts?
The anxiety around AI and its impact on white-collar work is understandable, especially with the rapid advancements in technology. Based on the most helpful WSO content, here are some key insights to consider:
AI's Impact on White-Collar Jobs:
Adaptation and Opportunity:
The Bigger Picture:
Mental Health and Perspective:
In short, while the future of white-collar work is uncertain, it’s not all doom and gloom. Focus on building skills that are hard to automate, explore industries with lower automation risks, and stay adaptable. The key is to view this as a challenge to evolve, not a dead end.
Sources: Calling out Goofy @CRE and the rest of Wall Street Oasis. Real Estate is the worst asset class going forward., Thoughts on AI and the future of the job market?, Work at top AM -- AI will virtually kill of entry-level roles in AM over the next 5yrs, Will robots replace your consulting or financial career?
“it seems the common sentiment is in 18 months all of this will be automated away by some Claude agent”
Common where? LinkedIn posts?
I have friends at FAANG and at YC-backed startups that have not written a line of code in years. Everything is AI. Claude Code... Codex... They literally haven't opened up an IDE in years.
I think it's amusing that people here think that they are protected from AI because they think AI can't make as polished slides as them or somehow can't do an LBO in Excel. I assure you... it can.
We are approaching the point where AI is like autopilot for planes: it's not perfect and it has its limitations, but in many cases, it surpasses the reaction time of human pilots. In certain scenarios (e.g., low visibility landings), auto-land is required.
None of what you wrote means that all white collar jobs will be gone in 18 months. Your friends still have jobs. Pilots still exist
This is idiotic. No it own't automate away all white collar jobs, or even 20% of them. Whatever is automated away will be more than compensated by new jobs.
Or the AI bubble pops and white collar life goes back to normal.
Yes just like the internet went away when the dot com bubble burst…
If the AI "bubble" pops, we will be sent into a deep recession and life will not go "back to normal". The amount of money tied up in the AI infrastructure buildout leaves nearly everyone exposed in some form, whether through 401k's, pension funds, or entire countries via their sovereign wealth funds. Considering the magnitude of the buildout (larger than the canals, railroads and grid combined per WSJ), it would be a global catastrophe if its failure resembles anything like the 2008 bubble.
Just saying what I'm thinking. It could be catastrophic.
I don't think it's a bubble. The models are way too useful.
The vocal users on this forum are delusional. AI tools have gotten to the point now where most of what I am doing all day is checking the output of models. The errors I am finding are still present but dramatically less than 3-6 months ago. I then send my slightly edited and improved work to a VP who runs the deliverable through AI and gives me comments which I in turn run through AI to make the changes. At some point, maybe in the next 18 months, my role in its current form will no longer be needed.
I really don’t believe AI is just a junior role problem either, it’s a headwind to the entire industry’s business model as it commoditizes our products.
I gotta be honest man. I use Claude all the time for stuff. I have no idea wtf you’re talking about.
I exaggerate a little when I say all I do is check the outputs. In any role there will still be lots and lots of meetings, adhoc projects, etc.
In the execution/analysis part of your job where do you feel Claude falls short?
Idk if your model output is internal / client-facing or what, but if someone senior to me got wind of trusting Claude completely with a deliverable I would be shit on. Honestly, the time it takes to iterate (and I’m not a shitty prompter either) is greater than the time it takes me to do it myself. Ai is helpful to find random linked cell errors or build a complicated formula but that’s really the extent of it, can’t have it build something from start to finish
We have internal financial models that AI uses as a starting point and completes it with the client’s financials and proposed transaction. It’s not that I am trusting the AI output as a deliverable. I am saying that my role has shifted from building the model to verifying and editing a model that is 90% there. In January this year AI’s final product was maybe 50% there which I why I am alarmed by the pace of improvement and worried what my part in the deal process will look like in the future.
100% agree and the same thing is happening at my firm. Some seniors now use claude to review everything and send comments and then I send it to claude again to fix. Models got to a point in the last few months where the output is very similar to what an analyst can deliver.
ok so, I did intern this summer and here are my 2c:
is the AI models any good? yes.
will banks continue to use AI? no. The token cost is too high and it will become economically unviable (as I found out in my firm this summer when they started to remove certain AI tools silently)
Maybe focus a bit more on grammar before you graduate, Cletus
Blue collar jobs aren’t safe either. Enjoy time with family and friends. We are very clearly in the singularity. AGI is imminent.
Remember to take your meds tonight
Funny how oblivious the typical finance professional is. Most haven’t taken the AI-pill.
AI isn’t a tool. We are building digital minds that will soon surpass the cognitive abilities of the human mind. The labs already have the recipe to automate all white collar tasks through sufficient RL. The AI researcher consensus is that AI r&d will be automated within the next 2-3 years triggering an intelligence explosion.
AI is a tool.
Similar to you for swallowing the propaganda in that way.
Sounds like you've OD'd on the lesswrong pill
Indeed. You are spot on. I was talking to my software engineering friends at FAANG and at startups, and I was shocked when they said they haven't written a line of code in years: and no, I don't mean they let the AI handle the easy stuff and they review it — I mean, literally, they don't write code anymore.
My current job as Copilot: it's able to read everything from Outlook, Teams, Excel, PowerPoint. I lot of the analysis I do in Excel now is also through Copilot. It is faster than me and it doesn't make any mistakes. If two analyses are spitting out different results, I ask Copilot to reconcile the two and explain to me the differences in methodology. If I'm not sure which methodology is better, I ask Copilot for the pros and cons of each. If I'm not happy with the formatting, I tell Copilot to make all hard coded cells blue and linkages green. It takes Copilot 30 seconds. If I'm presenting to seniors, I tell Copilot to slide up the results into a handful of PowerPoint slides.
There is nearly nothing AI can't do, with the exception of human relationships (e.g., originating deals).
“We are building digital minds . . .”
That’s a philosophical presupposition that what AI does is categorically similar to human thought, and that therefore its ability to replace human thought is based on how “good” it gets.
We ultimately don’t really know that that’s true (see Chinese Room thought experiment). AI “thought” could be an entirely different category of thing from human thought, and therefore unable to replace it. This is further supported by a few empirical observations of patterns of AI outputs being radically different from human outputs (e.g., studies showing AI text outputs being radically predictable vs. human widely dispersed & creative writing outputs).
Ultimately, the burden of proof is on the AGI apologists, and I don’t find their case compelling.
Fucking excited
As many of you know, AI doomers love to talk about the Turing Test: AI has passed the Turing Test when its output is indistinguisable from that of a human.
I have something called the Travis Test: when AI can do a better job on a model than a hung-over midwit summer analyst, who I'll call Travis (because lets be real, it's often Travis), it has passed the Travis Test.
Currently, we're pretty far from that. AI does great on cookie-cutter models that follow a strict template, but there's always some kind of custom feature/view needed for the company in question and it takes way too much direction to get Claude to do it right. Travis the human takes frustratingly long too, but eventually gets it.
Expect what? Unfalsifiable fluff?
Every time you pop up, you have a lot to say until it comes time to make a statement that you might be forced to stand behind. Then you disappear
How does AI make money? How is the massive expenditure justified? When it comes time to start paying for this shit, how many people/firms stick around? How do hyperscalers deal with the lawsuits coming their way? How do they deal with political risk, with user risk, etc?
All you ever talk about is how the technology is going to change the world at some future date, in some vague and woolly manner. No one gives a shit about the empty promises anymore. Why are we spending trillions of dollars for a better version of Siri?
Make a claim about where the technology will be in 6 months, so we can revisit and evaluate your claims. Saying "in the future, things will be different!" is stupid.
Calling Claude Code "a better version of Siri" is wild work. Comically bad take.
Why did you tag me twice in this thread? I’m obviously here, in this thread, posting.
Scurry on back to 4chan or whatever AI model you use to generate the underage AI girls you beat off to.
Yes, a meaningful percentage of white-collar work will be automated. AI is no longer a cute "tool" that predicts tokens. The models have already far surpassed the average white-collar worker. At this point, the main thing left is integration, which will take a couple of years at most.
I’m speaking from experience. I can say with confidence that none of the roles I’ve worked in since graduating from college will require a human within the next few months, and these were highly technical, critical-thinking-heavy roles. Fable 5 or Astra 6 can already handle these jobs end-to-end in a single session.
Want to make a friendly wager on white collar employment over the next 12 months?
What's the wager?
I for one love reading opinions on AI from people who know nothing about AI or what's happening in the industry. Keep 'em coming, always a good laugh.
What is your opinion?
From my experience with enterprise AI, I’m not too concerned with the impact of AI of white-collar jobs. The impact will be intense for sure but the metered-approach to enterprise adoption is a very real dampener. Cost-per-token has plummeted and will likely get cheaper but the ability to predict usage volume is very difficult. For my own employer, we’ve been extremely hesitant to do anything with token-based agent outside of very specific and narrow use cases and even then, there’s been far more debate on whether it’s worth the variable costs or not compared to actually adopting an agent. If this metered business model gets changed to flat fee or something like that, across the industry, then I think it’ll be much more catastrophic for the white-collar world.
AI won't just completely make grunt level work a commodity but instead it increases the output per capita for that type of work. Additionally you'll have senior level workers realize they won't have to delegate certain banal tasks. Rising up through an organization becomes different rather than easier: with higher output per capita, fewer junior seats are needed, and the technical (albeit repetitive and easy to learn) tasks that used to serve as the proving ground are allocated to a bot that can do said task in a fraction of the time. The ladder gets shorter and steeper, and the people who climb it are the ones who develop judgment without the repetition forcing it on them. More will have to learn how to sell themselves within org, products etc.
That being said, more autonomous opportunity opens up down market for above reasons
After some reflection, I have a more balanced take. Until AI can do the work of an MD or executive, someone has to fill those seats, and that means there will still be a pipeline of junior and mid-level talent. The first few years spent in standardized models and polishing CIMs are how you earn the judgment for the roles above. The junior role will probably change shape, but I'd bet it doesn't disappear. And if AI can win over clients and make final calls with real intuition, we have a much bigger problem than job replacement.
yes except now you need like 1/10 the juniors. Models are already a lot better than many juniors and models don't cry if you give them feedback like the zoomers.
I guess we’re all gonna be giving out happy endings in a shitty massage parlor in lower Manhattan in 5-10 years 💀
Who's gonna be paying for them 😭
Mamdani
I think counterintuitively the world is going to return to a relationship driven state as our trust in AI will slowly erode. And we will return to the long lost glory of cigar lounges, strip clubs, friday scotch and raw unfettered capitalism.
Not my original point but rather one from the hellsite known as x.com. AI is going to push everyone offline. The online world will become nothing but spam. Spam everywhere.
Lot of hilarious comments in this thread from people who clearly haven't touched a strong model. If you aren't using the latest frontier stuff you are doing yourself a disservice. Likely it can do most of your non IRL work at decent clip if you give it context and feedback. Stop using free models and work models!
That said I don't see it automating all of white collar work. It won't be able to if for example most banks and lots of industry never adopt it as seen from this thread lol. Intelligence isn't the constraint in most domains.
Astounds me as well. One can make the business/financial case for the leading Labs, but to question the technology itself is absolutely moronic. Convinced most of the anti-AI boosters on this site are either retards, or projecting their distaste of the Silicon Valley booster AI bro often making AI superintelligence/end of the world statements, on the technology itself. It isn't God incarnate, but is easily the most fascinating technological innovation we've had in a long, long time, and will go up there as fundamental changer of the human condition, in a few decades.
easily explained by the fact that many big industries are not letting their employees use the newest models. if their only experience is with the free tier chat gpt then of course not going to think it is even that smart.
"It is difficult to get a man to understand something, when his salary depends on his not understanding it."
Go to an art school. Or pick up a writing class. How about art history, philosophy, or even studying the classics. No kidding, I knew a guy whose dream in college was to become a classical Latin scholar and hang around Roman archeology sites and he's never gonna run out of a job.
Why? Because people always want to hear about stories of people from other people.
Lol your premise is hilarious. AI (LLMs) are not going to automate away 50% of the white collar workforce or whatever the number is
A single-digit % of jobs maybe automated away over time (really low value added work). But FAR overcompensated by new jobs and companies ploughing these savings into more employees to drive growth. Existing well run companies probably grow HC slower than sales (which is a far cry from massive unemployment). Lot more companies will be started
All of these changes will take TIME to percolate, it's not a 1yr story. Looking at a decade+. The real world isn't some frictionless vacuum
The tech is cool and useful, but despite some of the delusional posters that think 'muh we've reached singularity and what's the point of life', it's not going to utterly disrupt everyone's lives within the next 1-2yrs
People were saying similar things when ChatGPT first came out and we’re still here.
Interesting how the narrative is going from "AI is a useless bubble" to "AI is too dangerous. We need to regulate it."
Do you really not understand how bubbles endanger the economy or how companies try to use regulatory capture to crystalize monopolies?
On a finance forum?
Et aperiam id nulla aperiam in a. Est provident est eligendi ut doloribus magnam.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...
Ut non a nulla doloremque non. Necessitatibus maiores quas minus explicabo earum accusamus laudantium ipsum. Quo occaecati ea veritatis iure. Et dolores doloremque consectetur consequatur sunt molestiae. Iure esse non veritatis rerum sed provident natus. Et harum dolore sed ea velit animi ut.
Occaecati consequatur ratione natus magnam maiores in rerum. Adipisci expedita ducimus modi itaque voluptatem tempora corrupti. Quod quia similique ut repellendus eum. Odit est dolor totam molestiae. A quia dolores reiciendis omnis. Modi consectetur consectetur inventore laudantium.
Odit accusamus omnis nobis iste. Dolor harum molestiae ratione debitis. Repellendus tempore praesentium blanditiis deleniti. Quia autem sit labore quia magni. Aliquid ullam est nihil voluptas sit.
Minus dignissimos qui in possimus ea et. Laudantium aliquam eos dolores qui. Est est optio voluptatem iste eaque sapiente.