I’m surprised, hacker news is not questioning this in the slightest?
Is anyone really buying they laid off 4k people _because_ they really thought they’d replace them with an LLM agent? The article is suspect at best and this doesn’t even in the slightest align with my experience with LLMs at work (it’s created more work for me).
The layoff always smelled like it was because of the economy.
Hmm, actually lines up for me at least. It was a pretty big news item a few months ago when Salesforce did this drastic reduction in their Customer Service department, and Marc Benioff raved about how great AI was (you might have just missed it):
I’m beginning to doubt very much that will happen. AI/LLMs are already based on 99% of all accessible text in the world (I made that stat up, but I think I’m not far off). Where will the additional intelligence come from that SalesForce needs for the long tail, the nuance, and the tough cases? AI is good at what it’s already good at - I predict we won’t see another order of magnitude improvement with all the current approaches.
Hmm, am no LLM expert, but agree with you that the models themselves for the individual subject domains seem like they're starting to reach their peaks (Writing, solving math, coding, music gen...) and the improvements are becoming a lot less dramatic than couple of years ago.
But, feel like combining LLM's with other AI techniques seems like it could do so much more...
... As mentioned, am no expert, but seems like one of the next major focuses on LLM's is on verification of its answers, and adding to this, giving LLM's a sense for when its result are right or wrong. Yeah, feel like the ability for an LLM to introspect itself so it can gain an understanding of how it got its answer might be of help if knowing if its answer is right (think Anthropic has been working on this for awhile now), as well as scoring the reliability of the information sources.
And, they could also mix in a formal verification step, using some form of proof to prove that its results are right (for those answers that lend themselves to formal verification).
Am sure all this is all currently being tried. So any AI experts out there, feel free to correct me. Thanks!
The idea of formal verification works great for code or math where clear rules exist, but in customer support, there is no formal specification. You can't write a unit test for empathy or for "did we correctly understand that the customer actually wants a refund even though they're asking about settings." This is the Neuro-symbolic AI problem: to verify an LLM answer, you need a rigid ontology of the world (Knowledge Graph or rules), but the real world of customer interaction is chaos that cannot be fully formalized
Ah yes, and actually, Agreed (as mentioned, formal verification is only possible for "those answers that lend themselves to it").
Interesting that you mentioned Knowledge Graphs, haven't heard about these in a long time. Just looked up "Commonsense knowledge" page on wikipedia and seems like they're still being added to. Would you happen to know if they're useful yet and can do any real work? or are good enough to integrate with LLM's?
I mean, this might be a case where it’s actually sort of credible. It was a _very_ deep cut (basically half the workforce), the salesforce guy is a particularly over-the-top ai true believer, and if they are now reversing course and re-hiring, well, nothing has happened to the economy in the last couple months that would suggest that, if it was related to the economy. If anything, things are looking even more uncertain/ominous.
weird - even if AI was literally omnipotent and omniscient, you would still be bottlenecked on human's ability to actually evaluate and verify what it is doing and reconciling that with what you wanted it to do. Unless you're of course, willing to YOLO the entire company on output you haven't actually checked yourself.
for that reason alone humans will always need to be in the loop. of course you can debate how many people you need to the above activity, but given that AI isn't omniscient, nor omnipotent I expect that number to be quite high for the foreseeable future.
one example - I've been vibe coding some stuff, and even though a pretty comprehensive set of tests are passing, I still end up reading all of the code. if I'm being honest some of the decisions the AI makes are a bit opaque to me so I end up spending a bunch of time asking it why (of course there's no real ego there, but bare with me...), re-reading the code, thinking about whether that actually makes sense. I personally prefer this activity/mode since the tests pass (which were written by the AI too), and I know anything I manually change can be tested, but it's not something I could just submit to prod right away. this is just a MVP. I can't imagine delegating if real money/customers were on the line without even more scrutiny.
>weird - even if AI was literally omnipotent and omniscient, you would still be bottlenecked on human's ability to actually evaluate and verify what it is doing and reconciling that with what you wanted it to do.
one would hope that one ability of an 'omniscient and omnipotent' AI would be greater understanding.
When speaking of the divine (the only typical example of the omniscient and omnipotent that comes to mind) we never consider what happens when God (or whoever) misunderstands our intent -- we just rely on the fact that an All-Being type thing would just know.
I think the understanding of minute intent is one such trait an omniscient and omnipotent system must have.
p.s. what a bar raise -- we used to just be happy with AGI!
Genies, maybe? They are omnipotent and (generally) sufficiently aware of your desires that they shouldn't actually get "confused". Genies are tricksters that will do their absolute best to fulfill the letter of your wish but not the meaning.
Right, but you have to do a lot of work, and really most of your work is in this area. Less on the actual building stuff.
Figuring out what to build is 80% of the work, building it is maybe 20%. The 20% has never been the bottleneck. We make a lot of software, and most of it is not optimal and requires years if not decades of tweaking to meet the true requirements.
> In reality, even an ASI won’t know your intent unless you communicate it clearly and unambiguously.
I recently came to this realization as well, and it now seems so obvious. I feel dumb for not realizing it sooner. Is there any good writing or podcast on this topic?
Not really a bar raise - many people have assumed that "AGI" would mean essentially omnipotent/omniscient AI since the concept of the technological singularity came into being. Read Kurzweil or Rudy Rucker, there's a reason this sort of thing used to be called the "rapture for nerds."
If anything I've noticed the bar being lowered by the pro-AI set, except for humans, because the prevailing belief is that LLMs must already be AGI but any limitations are dismissed as also being human limitations, and therefore evidence that LLMs are already human equivalent in any way that matters.
And instead of the singularity we have Roko's Basilisk.
The problem goes deeper: verification is harder than generation. When writing an answer yourself, you build the logic chain from scratch. When verifying AI, you have to deconstruct its logic, cross-reference facts, spot hidden hallucinations, and only then approve. For complex cases (which are exactly what the humans were left with), the time for quality verification approaches the time to write from scratch. If the time becomes roughly equal, the AI stops being an accelerator and becomes just a source of noise that yields no productivity gains
>you would still be bottlenecked on human's ability to actually evaluate and verify what it is doing and reconciling that with what you wanted it to do.
this sort of assumes that most humans actually know what they want to do.
It is very untrue in my experience.
Its like most complaints I hear about AI art. yes, it is generic and bland. just like 90% of what human artists produce.
> even if AI was literally omnipotent and omniscient, you would still be bottlenecked on human's ability to actually evaluate and verify what it is doing and reconciling that with what you wanted it to do
no no no you don't get it, you would have ANOTHER AI for that
fair. I used to think that too, but I find at least for golang, the sota models write tests way faster than I would be able to. tdd is actually really possible with ai imo. except of course you get the scaffolding implementation (I haven't figured out a way to get models to write tests in a way that ensures the tests actually do something useful without an implementation).
Your final sentence is interesting. I'm not a strict doctrine adherent, but in TDD, don't you write some minimal test, then implement the system to pass the test?
yes, but I find it hard to constrain it to a minimal implementation. what usually happens is it writes some tests, then an implementation, and then according to the thinking, makes some modification. it works with a relatively precise prompt, but starts to go a bit off the rails when you say things in broad terms ("write tests to ensure concurrency works, and the implementation to ensure said tests are correct")
Move fast and break things. When a black box can be blamed, why care about quality? What we need is EXTREMELY strict liability on harms done by AIs and other black box processes. If a company adopts a black box, that should be considered reckless behavior until proven otherwise. Taking humans out of the loop is a conscious decision they make therefore they should be fully responsible for any mistakes or harms that result.
It's not even about humans "needing" to be in the loop, but that humans "want" to be in the loop. AI is like a genius employee who has no ego and no desire to rise up the ranks, forever a peon while more willful colleagues surpass them in the hierarchy.
Until AI gets ego and will of its own (probably the end of humanity) it will simply be a tool, regardless of how intelligent and capable it is.
Humans need to be in the loop for the same reason other humans peer review humans pull requests: we all fuck up. And AI makes just as many mistakes as humans do. They just do so significantly quicker.
It is impossible to verify anything in this article. For example "In recent internal discussions and public remarks". Where are these public remarks? How did this author get access to internal discussions? I rate this article as clickbait nonsense.
It does seem like Salesforce relies on Agentforce and therefore doesn't need as much support stuff. But the pressure was also to “reduce heads”, which is a bit of a tone-deaf way to describe firing thousands of people.
What is this site? maarthandam.com? Is it a blog? An AI generated “newspaper”? An internet Newspaper? The menu doesn’t work on mobile, no articles appear to have a by-line, and there’s no link to outside sources to indicate the provenance of these quotes.
For an AI agent to do a good job at customer support, you would need to
1. literally document everything in the product and keep documentation up to date (could be partially automated?)
2. Build good enough search to find those things
3. Be able to troubleshoot / reason / abstract beyond those facts
4. Handle customer information that goes against the assumptions in the core set of facts (ie customers find bugs or don’t understand fundamental concepts about computers)
5. Be prepared to restart the entire conversation when the customer gets frustrated with 1-4 (this is very annoying)
Point 1 (document everything) is the utopia that killed the project. In any complex system, documentation is a lossy compression of reality. The actual truth about how to fix bugs doesn't live in Confluence; it lives in senior heads, Slack chats, and intuition, and AI has no access to this layer of tribal knowledge
> declining service quality, higher complaint volumes, and internal firefighting
LLMs are a great technology for making up plausible looking text. When correctness matters, and you don't have a second system that can reliably check it, the output turns out to be unreliable.
When you're dealing with customer support, everyone involved has already been failed by the regular system. So they're an exception, and they're unhappy. So you really don't want to inflict a second mistake on them.
The counter: the existing system of checks with (presumably) humans was not good enough. For the last 15 months or so, I have been dealing with E.ON claiming one thing and doing another, and had to escalate it to the Ombudsman. I don't think E.ON were using an AI to make these mistakes, I think they just couldn't get customer support people to cope with the idea "the address you have been posting letters to, that address isn't simply wrong, it does not exist". An LLM would have done better, except for what I'm going to say in the counter-counter.
The counter-counter, is that LLMs are only an extra layer of Swiss-cheese: the mistakes they make may be different to human mistakes or may overlap, but they're still definitely present. Specifically, I expect that an LLM would have made two mistakes in my case, one of which is the same mistake the actual humans made (saying they'd fixed everything repeatedly when they had not done so, see meme about LLMs playing the role of HAL in 2001 failing to open the pod bay door) and the other would have been a mistake in my favour (the Ombudsman decided less than I asked for, an LLM would likely have agreed with me more than it should have).
If anyone has doubts about this, check out this map in another article on this site, it's second map in this article and it has this title under it:
"Potential Side Effects of a 100% Tariff"
They shouldn't have tried to force LLMs into doing something current models aren't designed for: semantic understanding of "unknown unknowns". Tier-2/3 support isn't just about picking an answer from a knowledge base; it requires deduction, empathy, and finding solutions that don't exist yet. Models excel at generating relevant text for FAQs, but the moment a task requires understanding novel context, correlating non-obvious facts, or recognizing subtle emotional cues from a customer, current LLM architectures fail ruthlessly
I'm aware that "what does Salesforce actually do?" is a joke but I also really don't know what they do and this article didn't help. They... have conversations with customers? What does the AI do?
They make hideously complicated software to help businesses manage their business. You need consultants to help integrate it and to make any changes to it. The interfaces are convoluted and require learning how they work rather than having any kind of discoverability. Switching to their systems often involves a dip in customer satisfaction. Switching off of their systems is nearly impossible by design.
A big chunk of it is like an enterprisey, old TwentyCRM. It connects with everything, and nobody got fired for choosing salesforce. And the decision makers all play golf together.
The most stupid narrative ever. If AI is so good for productivity, why don't you use it to make your 4000 workers produce even more than other companies? Why you need to fire them, so now you have hands tied to your back, and go back to produce the same amount of software? It is completely obvious that the goal is to fire workers, not to get AI stuff done.
There isn't an endless supply of features waiting to be built and money waiting at the door to pay for them. Do we really think that the only thing keeping them from being the biggest company on earth is their shortage of developer talent?
So you really believe that we arrived to the end of software? It's obvious that a competitor could create a better software (if that was possible with AI).
Salesforce is B2B and a complex software. I wouldn’t expected them to layoff that much support. Surprising. They should be empowering their support staff with AI tools to improve customer experiences.
Though I’m a bit surprised they have that much support staff.
Executive compensation is justified by "...enormous impact leadership decisions have on company outcomes..." yet when those decisions blow up spectacularly, the accountability somehow evaporates.
If your pay is 400 times average employee salary because of your unique strategic vision, surely firing 4000 people based on faulty assumptions should come with proportional consequences?
Or does the high risk, high reward, philosophy only apply to the reward part?
We all know the answer. There is no actual defense of inflated CEO salaries. It’s just the people in power maintaining their power and always has been.
In this case I think it came from the very top down — Benioff has been very bullish on AI and they’ve pretty much re-branded behind their Agent Force offerings.
Also probably a part of their go-to-market strategy. If they can prove it internally they can sell it externally.
Somebody has to be the brave experimenter that tries the new thing. I'm just glad it was these folk. Since they make no tangible product and contribute nothing to society, they were perhaps the optimal choice to undergo these first catastrophic failed attempts at AI business.
While someone does have to be the first to experiment I think you've implied a bit of a false dichotomy here. Experimentation can be good for sure, but it also doesn't have to involve such extremes. Sucks for the people left who now have to make up for the fact that someone's experiment didn't work out so well.
I think that as an employee it’s good to have a clear failure case study to point to from a large and credible organisation that this idea your boss has to fire everyone and just LLM everything isn’t going to work the way you expect it to.
The more examples of this going badly we can get together the better.
I think it was mostly a branding exercise, Salesforce wanted to signal to its customers that they are on top of this whole AI thing and there is no need to go to some unknown AI startup to "AIfy" their business. So they wanted to capitalize on FOMO / fear of being disrupted while using a bad labor market to improve profitability. They succeeded in this and made news around the world, but maybe not so many new customers.
Makes no sense - why would Salesforce's customers care if the company is using AI or not, other than when it impacts them (the customer) such as worse customer service.
This just seems a poor decision made by C-suite folk who were neither AI-savvy enough to understand the limits of the tech, nor smart enough to run a meaningful trial to evaluate it. A failure of wishful thinking over rational evaluation.
If you consider the extent to which our economy has become financialized, then you see these decisions have little to do with providing a product for customers but rather a stock for investors.
they contribute very little except of course that without jobs their products have created 14.8647% of US population would starve to death. HN seems like a perfect place where people upvote stupid shit like some of the most successful companies in the history of mankind contributing nothing to society. bravo!! :)
A bold statement. Who knew so many US citizen owed their food to an internet company! And not even Google or Amazon. Seems a reach, by maybe two or three decimal places.
He also uses cultural revolution tactics and uses the young ones against the old. I imagine AI house of cards will collapse soon and he'll be remembered as the person who enshittified Windows after the board fires him.
And when they can't undo their mistake will they accept the consequences, or will they cry to the government that there are no workers available to do the jobs so national policy must be modified to give Salesforce an even larger firehose of candidates to ignore? Companies complain endlessly that there isn't a huge stable of unicorns for them pick and choose from but those 4000 experienced staff were known good workers and they dumped them anyway to chase fantasies. Salesforce will demand the government fix their mistake for them. The larger the company, the more they expect to never have to pay for their mistakes.
Salesforce regrets firing 4000 experienced staff and replacing them with AI
December 25, 2025
New Chennai Café Showcases Professional Excellence of Visually Impaired Chefs
December 22, 2025
Employee Who Worked 80 Hour Weeks Files Lawsuit Alleging Termination After Approved Medical Leave
December 21, 2025
UPS Sued for Running Holiday Business By Robbing Workers of Wages
December 18, 2025
This Poor Man’s Food is A Nutritional Powerhouse that is Often Ignored in Tamil Nadu
October 5, 2025
Netizens Mourn as Trump Was Found Alive, Promising Tariffs Instead
August 31, 2025
Yes this reads like vacuous AI slop and and the **randomly bolded** text everywhere is a **dead giveaway**. At this point it's becoming a stronger signal than em-dashes.
> “We assumed the technology was further along than it actually was,” one executive said privately, reflecting a growing recognition that AI performance in controlled demonstrations did not translate cleanly into real-world customer environments
Yeah I can't see a source for the internal admissions of regret.
If we take out the AI part of this and treat it like any other project, if what they admit is true, it represents a massive failure of judgement and implementation.
I can't see anyone admitting that in public, as it would probably end their career, or should do at least. Especially if a company is a "meritocracy"
every single HN comment on these articles makes me doubt both the sentience of my fellow nerds and whether there are any actual human users of this website remaining.
Hacker News can only be good if enough people make the effort to make it good. There is always going to be a mix of things that push the standard up and things that drag the standard down. That's how averages and distributions work.
Unfortunately what we see from you is a pattern of low-effort comments, some of which don't even bother with basic sentence formation features like capitalization at the start and a period at the end. That's a high-signal hallmark of low-effort comments. Looking down your comment feed we see many single-line comments that are low on substance and high in snark.
The guidelines make it clear we're trying for something better here. They ask us to be kind, and to avoid snark and swipes. They ask us to converse curiously. They ask us not to fulminate, and not to sneer, including at the rest of the community.
It's fine to want HN to be better. As moderators we certainly do; that's why we do this job. But it requires us all to actually make the effort to be better in our own conduct. When you see comments from other users that aren't up to standard, we need you to use the tools that have always been here, like downvoting, flagging and emailing us (hn@ycombinator.com) so we can take action.
It isn't other people's job to make good enough for you whilst you conduct yourself in this way. If you really want HN to be better, please do your part to raise the standards rather than dragging them down further.
If I were you I would be more concerned with the fact that you have allowed what was once a well-respected forum to become little more than a spam platform for AI shills. You can silence me, but I am not wrong and I’m not the only one who has noticed this. It’s very obvious.
You should understand that one way people improve the standards of a commons is by imposing social controls on those who violate norms which create a healthy society, such as by shilling. That is normal behavior on every forum I’ve ever seen.
When you allow there to be 100x more of this mindless slop than of anything else, the most any individual can do to resist the tide is to contribute to the voices trying to make antisocial behavior come with a cost.
It works, and because it works, people will continue to do it until you figure out how to keep a clean commons.
PS. I suppose you would probably say the same thing to Rob Pike (if he were a user of your site which he doubtless is not).
Please don't sermonize to distract from your own record of disrespect towards HN and its guidelines.
The people you claim have “allowed” this have maintained HN for many years – 13 in dang's case, the majority of its history. The primary reason this is a place where people want to participate is because of the guidelines that have been developed and refined since HN's inception, and that we spend hours each day upholding. People have been heralding the decline of HN since it was barely more than a few months old [1], yet it continues to grow as a place where people want to showcase interesting work, which is what we most care about.
Generated comments and posts are banned, and we state this frequently. I spend time each day evaluating submissions and Show HNs to determine whether they're human-authored or AI-generated. We welcome people to flag generated content and email us so we can ban accounts with a pattern of posting it. Yes, it takes time for these mechanisms to kick in. HN is a public, anonymous site. Anyone can post anything, and the immune system takes time to do its work. That's always been the case.
There is a cohort of community members who have demonstrated a commitment to making HN better over several years through: (a) submitting good articles, (b) posting thoughtful comments, (c) observing the guidelines, (d) flagging bad submissions and comments, and (e) emailing us to point out guidelines breaches and to discuss the healthy functioning of the site. These are the people we listen to when they express concerns about HN's health, because they've established a track record of genuine contribution and care over several years.
From you, we see two comments prior to 2023, and little or none of the above kinds of actions. Instead: ragey fulmination, hyperbole, and ascribing views to us without basis. And now you hold yourself up as HN's heroic defender, having never undertaken the earnest, unglamorous, unseen work that other community members do to make this the place you claim needs you to defend.
Please, if you really want HN to be better, you are most welcome to start doing the things that other community members quietly do every day to help make it better.
I wanted to express similar sentiment, but I didn't understand how I would without leaving a rule breaking comment.
It's my sincerely held opinion that we're fostering a culture here that ignores the "human impact" of the technology that we're rushing to adopt.
I'm well aware that many members of this community have achieved "success" through software. This includes the rapid adoption of new computing paradigms, new technology stacks, new frameworks, etc.
I am fortunate to be employed. But around me, when I step out of my house, it's painful. People are hurting. They're unemployed. They're depressed. And the younger generation is even worse. They can't even afford to dream.
I live in a corporate world of forced smiles and fake enthusiasm. I would hate for that same culture to take root here. We need to be able to express significant doubt, or even cynicism against AI, without fear of backlash.
Competent management would have implemented a trial run to evaluate the feasibility of the plan. These sociopaths ensured their own failure by lunging for the prize without realizing they stepped off a cliff.
Is anyone really buying they laid off 4k people _because_ they really thought they’d replace them with an LLM agent? The article is suspect at best and this doesn’t even in the slightest align with my experience with LLMs at work (it’s created more work for me).
The layoff always smelled like it was because of the economy.
reply