How many jobs have they actually 'eaten'? For now, we mostly have AI labs and those in proximity claiming some number of jobs will be obsolete, but the actual job loss has yet to manifest.
When he retired a few years ago, most of that was gone. The attorneys and paralegals were still required, there was a single receptionist for the whole office (who also did accounting) instead of about one for each attorney, and they'd added an IT person... but between Outlook and Microsoft Word and LexisNexis and the fileserver, all of those jobs working with paper were basically gone. They managed their own schedules (in digital Outlook calendars, of course), answered their own (cellular) phones, searched for documents with the computers, digitally typeset their own documents, and so on.
I'm an engineer working in industrial automation, and see the same thing: the expensive part of the cell isn't the $250k CNC or the $50k 6-axis robots or the $1M custom integration, those can be amortized and depreciated over a couple years, it's the ongoing costs of salaries and benefits for the dozen humans who are working in that zone. If you can build a bowl screw feeder and torque driver so that instead of operating an impact driver to put each individual screw in each individual part, you simply dump a box of screws in the hopper once an hour, and do that for most of the tasks... you can turn a 12-person work area into a machine that a single person can start, tune, load, unload, and clean.
The same sort of thing is going to happen - in our lifetimes - to all kinds of jobs.
I recall the mine water pump had a boy run up and down a ladder opening and closing steam valves to make the piston go up and down. The boy eventually rigged up a stick to use the motion of the piston to automatically open and close the valves. Then he went to sleep while the stick did his job.
Hence the invention of the steam engine.
Condensing the workforce as you describe risks destroying redundancy and sustainability.
It may work in tests with high performers over short dutations but may fall under over longer terms, with average performers, or with even a small amount of atrition.
Having cog number 37 pick up the slack for 39 doesn't work with no excess capacity.
Complete aside, just because you brought up this thought and I like the concept of it:
My mom trained professionally as a secretary in the 1970s and worked in a law office in the 1980s; at that point, if you were taking dictation, you were generally doing longhand stenography to capture dictation, and then you'd type it up later. A stenotype would've been a rarity in a pre-computer office because of the cost of the machine; after all, if you need a secretary for all these other tasks, it's cheaper to give them a $2 notebook than it is a $1,500+ machine.
My observation so far has been that executive leadership believes things that are not true about AI and starts doing the cost-cutting measures now, without any of the productivity gains expected/promised, which is actually leading to a net productivity loss from AI expectations based on hype rather than AI realities. When you lose out on team size, can't hire people for necessary roles (some exec teams now won't hire unless the role is AI related), and don't backfill attrition, you end up with an organization that can't get things done as quickly, and productivity suffers, because the miracle of AI has yet to manifest meaningfully anywhere.
AI alone can't do that, even if you make the weakest link in the chain stronger, there are probably more weak links. In a complex system the speed is controlled by the weakest, most inefficient link in the chain. To make an organization more efficient they need to do much more than use AI.
Maybe AI exposes other inefficiencies.
I worked on both - my skillset went from coding pretty bar charts in SVG + Javascript to configuring Grafana, Dockerfiles and Terraform templates.
There's very little overlap between the two, other than general geekiness, but thanks I'm still doing OK.
in addition SalesForce grew in employment size in 2025 AFIK and 4000 jobs are for them only around ~~5%, which means it's to small to be a meaningful metric if you don't fully trust what their press department does (and you shouldn't)
still I see people using modern AI for small productivity boosts all over the place including private live (and often with a wastely underestimate risk assessment) so in the best case it's only good enough to let people process more of the backlog (which otherwise would be discarded due to time pressure but isn't worthless) and in the worst case will lead to idk. 1/3 of people in many areas losing their job. But that is _without_ major breakthrough in AI, just based one better applying what AI already can do now :/ (and is excluding mostly physical jobs, but it's worse for some other jobs, like low skill graphic design positions)
And as software developers, it would be silly if we didn't think that businesses would love to find a way to replace us, as the software we have created did for other roles for the past 60 years.
Though like non-GAAP earnings & adjusted EBITDA, very few care. Those that do are often old, technical, conservative & silent type of investors instead of podcasters or CNBC guests. RIP Charlie M.
Alternatively though if the market is bad and there not launching as many new products or appealing to as many new customers, customer support may be a cost center you’d force to have “AI efficiencies”
Companies like IBM & Klarna have made news for reducing positions like these & then re-hiring them.
AI, like most tech, will increase productivity & reduce headcount but it's not there yet. Remember, the days are long & the years are short.
Customer support is the "big" thing for AI replacement but it's also the worst role to replace with an AI, since customer frustration can (and usually does) lead to spurned customers switching to competitors.
Salesforce fired 4000 humans and replaced them with a dumb chatbot that couldn't answer many questions beyond the very basic. Customers absolutely hated the chatbot and started rethinking their Salesforce spend, resulting in a number of customers reducing their spend or switching to another CRM provider entirely.
Unless all the competitors are also going the AI route for support, because they're all equally greedy.
That's basically what happened with the path from humans answering the phone to "press 1 to..." to "say 1 to...", or website chat support being replaced by non-AI bots.
I really can't see an argument against the notion that customer service has been steadily worsening as a result of companies preferring to invest less on labor.
- OCR eat a good chunk of data entry jobs,
- Automated translation eat a number of translation jobs,
- LLM have eaten quite a few tier I support roles.
I don't have numbers tho, maybe people are still doing data entry or hiring translators on mechanical turk.
Initially machine translation was way worse (by professional standards) than people assumed, essentially useless, you had to rewrite everything.
As time went on, and translation got better, the workflow shifted from doing it yourself to doing a machine pass, and rewriting it to be good enough. (Machine translation today is still just 'okay', not professional quality)
On the initially set rates 15 years ago you could eke out a decent-ish salary (good even if you worked lots of hours and were fast). Today if you tried to do the work by hand, you'd starve to death.
While they help with programming, I feel like the scope of my tasks have increased over time as well. I feel like this is happening to me - I'm insanely more productive and my tech stack has increased hugely over the past two years as has my productivity.
But I don't make significantly more money, or get a ton more recognition, it's just accepted.
"okay" is an overstatement. It's "readable" if you put in triple the effort needed to read proper native language. But you're doing a lot of work re-translating machine languae in your head to understand it.
I guess for businesses that's "good enough". Very few products ever truly get lambasted for bad localization.
The question is no longer whether AI will put people out of work, but how many and how quickly.
to be fair this positions never made that much sense as they tend to cause more trouble then they are helping on the long run, but they exist anyway
and companies should know better then throwing away "junior, not yet skilled, but learning" positions (but then many small startups also are not used to/have the resources to teach juniors, which is a huge issue in the industry IMHO)
but I imagine for many of the huge "we mainly hire from universities"/FANG companies it will turn into a "we need only senior engineers and hire juniors only to grow our own senior engineers", this means the moment to you growth takes too long/stagnates by whatever _arbitrary_ metric you get kicked out fast. And like with their hiring process they have the resources, scale, and number of people who want to work for them to be able to really use some arbitrary imperfect effective discriminatory metrics.
Another aspect is that a lot of the day to day work of software engineering is really dump simple churn, and AI has the potential to massively cut down the time a developer needs to do that, so less developers needed especially in mid to low skill positions.
Now the only luck devs have is that there is basically always more work which was cut due to time pressure but often isn't even supper low priority, so getting things done faster might luckily not map one to one to less jobs being available.
you get HR's glossy `Exit Packet' with a cover of a pristine chartered white catamaran on the translucent aquamarine Carribbean in a palm treed cove, afloat with bikini babes lounging on deck, five minutes to fill your cardboard box, and manhandled by two security wide-shouldered bulls squeezing your arms to your sides gripped under your forearms, marching you down the aisle with rubber-knecking wide-eyed heads watching you repeatedly slip, fall forward, the experienced bulls lowering your arms to regain your shoes' purchase with the carpet, hurriedly packing you into the elevator with silent ignominious stares and hushed whispers of those packed in around you, then past hot Tanya at Reception in front of every Tom, Dick, Irene, and Harry, out the main revolving glass door entrance, into the overcast dishwater grey and miserable wind blown wet in truth is Seattle, to the sidewalk, left at the curb -furthest from proper and successful glitterati as possible.
All double-time haste, signalling to everyone get this despicable loser/criminal POS off the property, fast.
Its over, you're done. Sooner than you thought possible, you're freezing in a tent wasting days into months: tailing-downward foodbank boxes, the TIDE(tm) pee-bottle, those el-cheapo dirty gloves with the tips cut off, dirty layered thriftstore fleeces, its under filthy roaring I-5 for you pal, and your ilk of dangerous insane addict-crazed zombies screaming outside your hideous blue-tarped tent, and where's the knife.
The answer to the fundamental question of your entire existence and net worth has reduced to simply ask: "How do I best empty my bowels"?
ICEstapo hunter/killers gunning for you, to flush you offshore unseen and forgotten forever. You better run, you better run your ass off. Why didn't you study harder all those wasted years?!
Welcome to Sam Altman's Club.
So jobs being killed by AI are basically being killed same way that office number crunching technology killed administrative assistant positions and put those tasks onto other people.
Take for example a purchasing department for a big company. Some project needs widgets. Someone crosses the specs against what their suppliers make. They take the result of that and makes a judgement call. AI replaces that initial step so a team of N can now do the work that formerly took a team of N + Y. Bespoke software could have replaced that step too but it would have been more expensive, less flexible, etc, etc. since there's all this work required to make human facing content into machine parsable content, including the user's input and the juice simply wasn't worth the squeeze. With AI doing all that drudgery on an as-needed basis the juice now is worth the squeeze in some applications.
And the sick thing is that the company that tries to be smart longer term won't be able to compete with the short term companies that cut as much as possible using AI to maximize short term benefits. Long term these 'cut everything' AI leaning zombie companies won't last, but they will last long enough to undercut and take out the longer term thinking companies with them.
AI is an entropy machine. It sucks all momentum from everything it touches.
Every task the AI can juggle for you is one you don't have to. If your department goes from 4 to 3 great, if it goes from 2-1 or 1-0 that's fine too. Companies already exist at those numbers. You can think about it in terms of "a company can now be bigger before they need a dedicated person/team for job X" if that helps take the emotion out.
Companies that can benefit from these sorts of purchasing use cases don't exist at 0. And at 1, this is already a minimal need so not useful. Past 2, and you need 2 staff anyways from reasons listed and more, so AI doesn't give a cost benefit, but it does introduce all kinds of risk (there's a lot more than just comparing SKUs) and weaken relationship with supplies, which can kill a company. Just doesn't make sense all around.
To think the same isn't happening all over the place and will only continue is ignoring just how powerful this tech is.
There is so much else that people do. So many details that are just being ignored because of short term 'gains' that justify ignoring so many details.
If this person plus a junior represented "1.3 engineering knots," he's saying... "actually, I'm still 1.3 engineering knots without him."
When this person leaves, they go find someone else who is 1.3 engineering knots. The junior represented .3, without the 1., it doesn't matter that much. Headcount strategy shifts.
So the company you talk about is already in the entropy vortex. It has no momentum. It has no future. It just hopes it can keep doing what it is doing now.
Their function is around reconciling utilization and bills from multiple related suppliers with different internal stakeholders. They do a bunch of analysis and work with the internal stakeholders to optimize or migrate spend. It is high ROI for us, and the issue is both finding people with the right analytical and presentation skills and managing the toil of the “heavy” work.
Basically, we’re able to train interns to do 80% of the processing work with LLM tooling. So we’re doing to promote two of the existing staff and replace 2/6 vacancies with entry level new grads, and use the unit to recruit talent and cycle them through.
In terms of order of magnitude, we’ll save about $500k in payroll, spend $50k in services, and get same or better outcomes.
Another example is we gave an L1 service desk manager Gemini and made him watch a YouTube video about statistics. He’s using it to analyze call statistics and understand how his business works without alot of math knowledge. For example, he looked at the times where the desk was at 95th percentile call volume and identified a few ways to time shift certain things to avoid the need for more agents or reduce overall wait times. All stuff that would require expensive talent and some sort of data analysis software… which frankly probably wouldn’t have been purchased.
Thats the real AI story. Stupid business people are just firing people. The real magic is using the tools to make smart people smarter. If you work for a big company, giving Gemini or ChatGPT to motivated contracts and procurement teams would literally print money for you due to the stuff your folks are missing.
This is to say that we know from looking at outcomes over the long term that the kinds of concrete gains you're describing are offset by subtler kinds of losses which most likely you would struggle to describe as decimal numbers but which are equally real in their impact on your business.
Public LLMs have been around for 3 years, and adoption is still nascent. We don't have any long term data, and the longest term data we have involves a bunch of outdated models. Most people are still awful at using LLMs, and probably the most skilled users are the college kids who are graduating right now (the youth is always god-tier with new tech).
I cannot think of anything more foolish right now than not trying to leverage SOTA models to save you money, especially because you heard rumors of shadow losses that can only be found in the bottom line.
AI is an entropy machine for everything it touches. Companies that runs off of AI is a zombie company. Without people to understand an industry, what does your company ad? Without people to see new revenue streams, new directions, and most importantly NEW RISKS to your business model, you are a dead zombie company living off what the previous living, growing company built. But does that matter to owners when they get $500,000 more a year in their pockets?
Gemini "analysis" looks superficially detailed but if you actually do any deep dive (or even a shallow dive) you're going to start finding all the things its wrong about because it's not actually analyzing anything; its just connecting most likely words together. Very crucial, because of the ways LLMs work, they can't generate any conclusion that isn't already in its data set. But it's exactly those sorts of non-obvious conclusions that are the reason you hire people to do Finance.
Will AI eventually take over these kinds of jobs? Sure, in 10 to 20 years when actual AI exists. But LLMs aren't AI. They're just brute-force black box machine learning algorithms.
Except it seems like the opposite is happening. CS grads have high unemployment. Companies laying off staff.
The rhetoric doesn't seem to add up to the reality.
Do you use tech to grow your business or increase dividends?
Also reducing staff via attrition shows far better management skills than layoffs which imo says more about the CEO & upper management.
- Translation. See: Gizmodo firing its Spanish translation team and switching exclusively to LLMs.
- Editors. See Microsoft replacing their news editors at MSN with LLMs.
- Customer service. Various examples around the world.
- Article graphics for publications. See: The San Francisco Standard (which used it for various articles for a period), Bleepingcomputer.com, Hackaday (selectively, before some pushback).
- Voice acting. The Finals game used synthetic voices for the announcer voices.