AI Will Shrink Amazon's Workforce In the Coming Years, CEO Jassy Says 36
In a memo to employees on Tuesday, Amazon CEO Andy Jassy said that the company's corporate workforce will shrink in the coming years as it adopts more generative AI tools and agents. "We will need fewer people doing some of the jobs that are being done today, and more people doing other types of jobs," Jassy said. "It's hard to know exactly where this nets out over time, but in the next few years, we expect that this will reduce our total corporate workforce." CNBC reports: Jassy wrote that employees should learn how to use AI tools and experiment and figure out "how to get more done with scrappier teams." The directive comes as Amazon has laid off more than 27,000 employees since 2022 and made several cuts this year. Amazon cut about 200 employees in its North America stores unit in January and a further 100 in its devices and services unit in May. Amazon had 1.56 million full-time and part-time employees in its global workforce as of the end of March, according to financial filings. The company also employs temporary workers in its warehouse operations, along with some contractors.
Amazon is using generative AI broadly across its internal operations, including in its fulfillment network where the technology is being deployed to assist with inventory placement, demand forecasting and the efficiency of warehouse robots, Jassy said. [...] In his most recent letter to shareholders, Jassy called generative AI a "once-in-a-lifetime reinvention of everything we know." He added that the technology is "saving companies lots of money," and stands to shift the norms in coding, search, financial services, shopping and other areas. "It's moving faster than almost anything technology has ever seen," Jassy said.
Amazon is using generative AI broadly across its internal operations, including in its fulfillment network where the technology is being deployed to assist with inventory placement, demand forecasting and the efficiency of warehouse robots, Jassy said. [...] In his most recent letter to shareholders, Jassy called generative AI a "once-in-a-lifetime reinvention of everything we know." He added that the technology is "saving companies lots of money," and stands to shift the norms in coding, search, financial services, shopping and other areas. "It's moving faster than almost anything technology has ever seen," Jassy said.
PHB*-to-English Translation: (Score:5, Funny)
"Sales are in a slump, so we'll fire employees yet blame staff shrinkage on replacement by our wonderful shiny AI"
* Dilbert's boss who is light on knowledge but heavy on buzzwords and BS.
You're underestimating the AI boom (Score:2)
There are tons of things in every single organization that CEOs have held back on automating because they just didn't believe it could be automated. Like the old joke, go away or I will replace you with a tiny Perl script.
AI has convinced every CEO on the planet that everything can be automated. Add to that that the current American administration is driving the
Re: (Score:2)
Whether llms really make a huge difference the constant press AI gets has convinced every single CEO that they need to automate everything. There are tons of things in every single organization that CEOs have held back on automating because they just didn't believe it could be automated. Like the old joke, go away or I will replace you with a tiny Perl script. AI has convinced every CEO on the planet that everything can be automated. Add to that that the current American administration is driving the country into a recession via reckless spending and trickle down tax cuts. So there will be heavy pressure to fire people with automation. And it doesn't even matter if it works. If it works great and if it doesn't we will be knee-deep in a nearly permanent recession so you'll just have to work 80 hours a week to make up for the broken ai. Like having a useless coworker but it's a machine.
You missed a step. You'll have to work twice as hard as today to make up for the broken AI, and you'll do it for far less pay than you have today, because by the time they get back around to hiring humans to fix the automation that failed, those humans will be so desperate for a wage that they'll work for pennies on the dollar compared to today.
Re: (Score:1)
Re: (Score:2)
While I agree with your sentiment, there is one problem. Amazon's revenues are increasing at a rate of about 10% per year, not decreasing. https://www.macrotrends.net/st... [macrotrends.net].
I'm Still Not Seeing It (Score:5, Interesting)
Yesterday I had a need to extract a bunch of documents our of a web based ERP system. The ERP system has a pretty extensive and quite good API. But, I wasn't in the mood to study and learn a new API to do a simple script that would pull all customer numbers and then iterate through each customer and download their files.
I decided to try the Vibe coding that the cool kids are supposedly doing. I tried Copilot, ChatGPT, and Claude. None of them could assemble even a basic remotely working Python or JavaScript script. Attempt after attempt, they al returned completely incorrect and completely non-functional paragraphs of code. I could get nothing of value from AI.
Today, I spent three hours studying the API and finding the relevant calls. I had a fully functional Python script running in 30 minutes and the task completed two hours later.
AI 0 Weak Programmer 1
I just don't see successful replacement of human coders with AI.
Re: (Score:2)
In Amazon's case, I'm having a hard time believing they'll replace people they can treat as slaves with machines that don't care how much you abuse them. I mean, really, how superior can you feel to a machine?
Re: (Score:1)
AI has a tough time with older languages (Score:2)
If the language is really old and unchanging like say c++ or c then you can get working code out of it because they're just isn't a lot of things that break completely under those languages. And if it's very very new you can get working code because the data it was trained on is very new.
What I found with python when I had to upgrade a python 2 app to Python 3 was any code I was giv
Ditto (Score:2)
Today, I spent three hours studying the API and finding the relevant calls. I had a fully functional Python script running in 30 minutes and the task completed two hours later.
AI 0 Weak Programmer 1
I just don't see successful replacement of human coders with AI.
Today's AI adventure....Claude...simplify my stupidly complex loop....it returned the exact same code...OK, fine...I know you can simplify it and lambda it...but I was feeling lazy and it'll probably pass a PR.....Claude, write me a unit test for my stupidly complex loop function (this is integrated into my IDE, so it has full visibility into the project)....the fucking thing doesn't compile. Why?...the braces didn't match!?!?!?....OK, I fix the braces for them...well, it hallucinated Mockito functions I'v
Re: (Score:2)
I wrote a unit test...assuming I was just being paranoid....nope...I had garbage on my hands...generated by Claude 4.0 and it LOOKED right, but wasn't....and it almost got past me.
Your boss needs to hear about that - seriously.
Re: Ditto (Score:2)
His boss's boss will probably say to fire him for not drinking the cool aid.
The think tanks that advise the companies boards are pushing the AI adoption very hard, so all the CEOs are complying, whether they believe it or not.
The individual contributors and middle managers onky have the choice to comply or get laid off, most of the time the latter.
Re: (Score:2)
Think-tanks are likely full shareholders then I guess.
AI is a religion & if I doubt, I must be purge (Score:2)
I wrote a unit test...assuming I was just being paranoid....nope...I had garbage on my hands...generated by Claude 4.0 and it LOOKED right, but wasn't....and it almost got past me.
Your boss needs to hear about that - seriously.
Facts don't matter because AI is a religion at this stage. Embrace the LLM and glory will come to you and your projects. If it worked, they'd be right...but it doesn't...so the smart ones think...OK, Claude 4.0 can't match braces or use the correct API, but Claude 4.1 will!!!! The ones with no experience think I'm an old guy....just like those fat grey beards I had to deal with when I was starting out who were telling me "the internet is a fad...it's only good for porn & selling books...only a moron
Re: (Score:2)
I have had almost no luck having AI work with proprietary, or even unpopular APIs.
Caveat: I have had much more luck with them when using derived AI trained on said proprietary APIs, assuming sufficient code exists to leverage. So perhaps you need to tell the boss you need to spend money to hire a team of people to train an AI on your various APIs. Once the bill for that one comes in, you'll be safe.
Re: (Score:2)
Nobody that actually looks and is willing to assess the results without prejudice is seeing it. My 2nd Semester coding students (engineering in a non CS field) are not seeing it. Their assessment was "often faster for search than looking at the docu, but cannot code for shit". And that is in Python where a lot of example code is out there.
Re: I'm Still Not Seeing It (Score:3)
Today, I spent three hours studying the API and finding the relevant calls. I had a fully functional Python script running in 30 minutes and the task completed two hours later.
... THOSE are the parts you're supposed to use AI with ... if the API docs are online 3o will do that much for you.
It's like you just posted on stackoverflow for a fully working solution then gave up because nothing you copy pasted works right. It's about the dumbest possible way to use an information tool.
Treat AI like an information index, just like SO, just like a search engine, it's that kind of tool. It's a very, very good index that sometimes points at exactly what you asked for but you still search,
Re: (Score:2)
Today, I spent three hours studying the API and finding the relevant calls. I had a fully functional Python script running in 30 minutes and the task completed two hours later.
... THOSE are the parts you're supposed to use AI with ... if the API docs are online 3o will do that much for you.
It's like you just posted on stackoverflow for a fully working solution then gave up because nothing you copy pasted works right. It's about the dumbest possible way to use an information tool.
Treat AI like an information index, just like SO, just like a search engine, it's that kind of tool. It's a very, very good index that sometimes points at exactly what you asked for but you still search, read, integrate, repeat. What you tried and failed to do, vibe coding, is like irresponsibly farming your work out to an intern. Go ahead and try that. You get the results you deserve and it doesn't make the intern useless, you're using them wrong. No, it doesn't replace you, that's not how you use it.
Right. We also don't even know what he did.
E.g. did he even give the LLMs the API documentation, or point them to it? Or did he just expect the magic 8 ball to spit out the right answer from the ether? Did he give the LLM a sample implementation, even if in some other language?
I see so many smug posts from people who don't even know that they are just saying "I don't know how to use LLMs".
Re: (Score:2)
I see lots of people blaming the user, but if you look at the hype and what so many of the influencers and what the CEOs etc are reported to be saying - he used it exactly right. The way it's promised to be able to be used.
If it's actually just a search tool, then it's not replacing any coders right? You still need people to do all the hard stuff - no one needed AI to make slightly different SO posts to copy paste and massage.
The other issue is - often to use it like you're saying, you need to somehow feed
Re: (Score:2)
If it's actually just a search tool, then it's not replacing any coders right?
Depends on what you mean.
E.g., I've given ChatGPT API documentation, and a sample implementation in one language, and asked it to give me a sample implementation in another language. And what it gave me worked, with minimal tweaking to use in my project.
Could I have done it myself? Sure. As fast? Heck no.
(And yes, I still did all the same code review, QA, etc. that I would have had to do anyway, even if I had sweated out every line myself.)
Re: (Score:2)
I'm willing to admit I'm probably an edgecase, but I asked AI - Claude 3.7 Sonnet in my case - to do the same sort of thing(translating one scripting language to another) and what it spit out didn't work at all (oh, it didn't error out, it just ended up doing nothing when run). It seems very context dependent.
I also asked it to do something very simple in bash - get the users of a group via LDAP and write a .5klogin file with them. I spent a couple hours trying to massage what it created and gave up. Now,
Re: (Score:2)
I'd argue, for me, the problem with AI coding is it's a lot like a slot machine. I have had it work once early on and I was quite pleased, but then each subsequent pull has failed. The issue is when I try and guide it out with suggestions or telling it what and how it failed, I end up spending a lot of time hoping it's going to work - but it mostly doesn't.
Yeah, it's very uneven.
There are some problem spaces where I am pretty confident I know what to use it for and what to bail out on early. Others where the jury is still out.
I like the slot machine metaphor! :D
Re: I'm Still Not Seeing It (Score:1)
I wish (Score:2)
I wish every employee at every company whose elites claim they can replace large swaths of people with AI could just walk away and leave those companies with no one left to run them. See how far they get with AI alone.
Not far.
Nope (Score:2)
Actually Amazon will shrink Amazon's workforce
Will AI replace Andy Jassy? (Score:1)
CEO admits that AI will replace human workers (Score:2)
The real news story is that CEO admits that AI will replace human workers. Before it has been all about "robots won't replace humans"
https://www.digit.fyi/robots-w... [digit.fyi]
https://www.livemint.com/Techn... [livemint.com]
https://www.aa.com.tr/en/world... [aa.com.tr]
Good! (Score:1)
All this means (Score:1)