Inside Amazon’s $100B Automation Gamble

26:37We're going to have AGI tomorrow. He was pushing back on that. I think the point that he really landed on here was in regards to agents, right? There's tons of venture capital hype around agents. Agents are being pitched everywhere is the next big thing. He basically said the industry is kind of overshooting the tooling with its present capability, which anyone who's kind of used these agents that are like claiming to be travel agents and book everything and do multi-steps for you knew already. And so it's kind of refreshing to have this very serious voice of technical leadership in AI speak to you honestly and say, yes, the agents are not quite there.

27:12That does not mean that AI is, sorry to the skeptics, all not going to work and won't ever get there. It's just the timelines we need to be a little more realistic on and we need to explore different technical angles to make these things work. And maybe we haven't figured those out yet. Well, the rise of agents was something super interesting to me when I was covering enterprise businesses at the Wall Street Journal, because you basically saw the trajectory of AI go from, you know, ChatGPT comes out, it's a consumer phenomenon, Microsoft builds their own search engine, and then basically co-pilot was the term at that period.

27:46These were assistants that sit alongside you and can make you more efficient and can make your job that much easier to do, help you write email, shit like that. And then with no major, in my opinion, technical breakthroughs between the co-pilot era, agents just sort of appeared on the scene. And suddenly these things that were like debatable as to how good they really could work became even more essential and autonomous in what they were supposed to do for you. And the technology didn't seem like it took a major leap. And so I was kind of scratching my head this whole time, seeing Microsoft specifically go from like, oh, we have this co-pilot tool and everyone should pay extra for Microsoft 365.

28:24So you can use their co-pilot to, oh yeah, now you're supposed to use agents where it does all the shit for you. Obviously you want to show you that you're an AI forward company because that you get a premium on your stock. And so you have to make all kinds of statements. But when it comes to utility of this shit, like they were just starting to figure out co-pilots. And so now you're going to start claiming there's agents. I just updated my Apple operating system, right? And if there's any company that's trying to force this AI thing that doesn't seem to have a view of what the actual value it drives, it's Apple, right?

28:54It promised, you know, all these phones based on Apple intelligence and they haven't really figured out what the value is. And it's like, you know, they mess with your contacts and they say, oh, here's who we think. But it's like I had a nice favorite list. It was organized. How is your like smaller, weirder version better than like my nicely organized one? And so, yeah, a lot of these companies have an incentive to force an AI narrative. I mean, Tom, you just wrote about Salesforce, which, you know, Mark Benioff has been like agent force, agent force, agent force. I mean, he's been the most sort of egregious in terms of just having, you know, the name of his company, AI, you know, he's a, he's a sales guy, you know, he knows how to try to associate, but that, that sort of heavy handed approach hasn't resulted in a good stock performance.

29:43No, certainly not because they're stuck in this position where, you know, you see the rise of these agents or AI companies. And it's like either someone like a sales force is going to be able to, you know, leverage off of that and be able to sell these tools to the customers, or there's going to be a whole new wave of tools built by not Salesforce that are just going to replace all of SaaS. And you've seen there'd be like a negative feeling about SaaS for like the last year or so because of that worry. But the knock on Salesforce is that their agents don't work, right? It's just like you're selling a product to people that have no real utility.

30:16It gets stuff wrong all the time. And I think that's what Karpathy was really trying to get at in his comments was like, look, there is a world in which agents have incredible impact on the economy and the way we do work and jobs and the labor grid and all of these things, but we're not there yet. And if you're selling it to people now with the claim that it's one tick away from AGI and one tick away from being able to replace a huge percentage of your workforce, you're bound to be disappointed. It's going to fuck everything up. And I don't know. What's interesting to me with his comments is how much of the hype around AI and the amount of money that Sam is consistently able to raise or stock jumps that he's able to generate through announcements are built on the premise that AGI is just around the corner.

31:04Like how much value is there in just a good utility for the next five, 10 years? Uh, if AGI just doesn't exist on the timeline that people like Dario have been claiming. I know we still have some of those Ed Zitron fanboys hanging around, just being like, uh, what is this podcast? I, I still, you know think the models themselves are so valuable i talked to chachi pt all day long like i is it not like a key part of your lives like i feel like it's extremely valuable on day-to-day questions relative to like google i i don't know um yeah you guys don't agree or no i use it i use it quite a bit i've used it for um helping me figure out like recipes and stuff and it's a much better search tool for a lot of things but it's not even search because it does generate for you what you need, you know, and provide things for you.

31:53And it helps me, you know, proofread and things like that. Um, I found it to be an incredibly useful tool, but I guess I do agree to, I think where Karpathy was getting at, which I think is where we land here is he, he was landing at how these can be great assistance and tools, and this is an evolution and they are better. But what the problem is, is you want it to make fewer, he actually wrote this in his post. I want it to make fewer assumptions and ask collaborate with me when not sure about something i want to learn along the way and become a better programmer not just get served mountains of code that i'm told works and i was like that's kind of the trough where we're kind of digging out of right now but that doesn't mean at all that these aren't useful tools right what do you want what do you guys make about uh dario and his claims because he's doesn't have an amazing track record right now of like timelines of when stuff is going to happen and the thing that i keep thinking about That was his claim, I think, a year ago that basically at this time now, more than 50 % of code would be generated by AI.

32:54And I don't know what the actual number is, but everyone was dunking on that because we're clearly not even close to that right now. It's far below that. Yeah. Yeah. And it reminds me of the self-driving era where there were also outlandish claims about the timeline it would take for this technology to be street ready. and they were way off. And I was super bearish about that, to be clear. At the time I was in Ubers in 2016, I was like, they're intervening all the time. We are not about to be in Ubers without... Well, Uber in particular didn't have a exact record on it. Like their tech was bad.

33:32But that's when they had the acquisition. Was it auto? Yeah. I mean, there was a period where Uber was really trying to assert that it was going to be the leader in self-driving. And I think, I mean, you know, and come at us, Ed Zitron, haters, or the haters that edgerton brought to the podcast but like i you know i think i'd like to say to them they're they're shooting us in the comments it's our fucking podcast we can talk we're supposed to talk like whatever they just want to hand over the mic to somebody else yeah anyway go on yeah get your piece in eric you're back yeah they're still commenting on that episode they can shit on me as much as they were just talking about ai bullishness or not i'm fine with the negative comments i love the comments that are like i can't believe they've left the comments open And it's like, yeah, we have thick enough skin that we can take a little bit of criticism.

34:18What I don't get is that on our own podcast, we're not supposed to – just because you know that guy and he's on our podcast, we're not just handing it over to him to just say whatever he wants the whole time. I don't know. I'd like to see the tail of the tape on that one. I like Ed and we've continued – We let him talk so much. Yeah. He had a lot of time. The time of possession was strong for Ed in that episode. I have no idea what they're talking about. And you would just hit into a new objection. Like if you're having like a through line argument, you need to hold some. It's like we're arguing about this thing.

34:48And then you make points around it. You just be like, oh, what about energy? You know? Oh, anyway, I'm still still having flashbacks. Look, anyway, I'll bring these haters back here and saying like, I think self-driving cars are really impressive. I think I do pay a premium in San Francisco if I can take a Waymo versus an Uber. And a lot of the promises. Yeah, the timelines are way off, but the technology did pan out to be. non-economical. There's all kinds of issues around how much it costs to run these things, but it proved itself at a different timeline than was initially predicted. And I guess the question is with AI, is there a timeline that does make sense that someone like Karpathy agrees with, where we do reach AGI?

35:28Or is this capability of these models where ChatGPT is now, which does seem like it's leveled off quite a bit, is that enough to meet the hype of something? It may not be to the level of displacement but like there's a lot of utility we're all seeing that in in these tools and like where does evaluation you know net out with that sort of thing and i don't think he has an answer to that but that to me is the question we should be asking in the next year or two it's hysterical though unlike self-driving cars our whole economy seems staying on this question at the moment self-driving cars were more of like on the side these companies can do this this was a fad and with this i do see it is right you know kind of maybe you got a little annoyed that it's like oh man these companies are getting profiles that act like they're you know geniuses when it like they can't do what they say they can do but like um now i think we're underrating self-driving cars but the but the yeah the ai stuff right now i mean just nvidia is the most valuable company you know in the world it's it so much of our economy hangs on uh continued adoption of artificial right and specifically capex spending on data center growth and that's like i would say 80 to 90 the responsibility of sam and striking all these deals that are basically you know deals upon deals of data centers to be built and so like yeah the point i guess you're making is like yeah it does need to be more than where it's at now like you're not gonna be building all these data centers because agi won't exist like that is that we'd like to see another leap you know it's like okay chat gpt has been great coding's cool but like i feel like our minds haven't been blown in a little bit right i mean we we get impatient when's the next video stuff i was gonna say sorrow was a big leap for the new sore app um it you can still which is still number one on the app store by the way i checked is it okay yeah i see i see the ones that make it over to tick tock um but i i don't love them like well there was um i will say this this deep this sort of video of like a a pastor you know condemning billionaires uh went pretty viral over this week that a lot of people fell for um including a lot of my friends yeah i like they are getting reality like i already hated the like i hate those the people love those like skits where it's like clearly these are actors but they get mad at it in the comments you know what i'm talking about where it's like yeah it's like a husband and wife are fighting or something but like all the comments it's like you know treat it as if it was like real where they're like this guy's a jerk or whatever you know it's like but these are clearly actors i feel like the sort of thing is like the next evolution of it i just hate the media consumer that like buys into the fiction like they don't they don't care like if it's real or not they're like oh it feels real and so i'll just react to the characters because it's like wouldn't it be great if you know this is happening There's been a, there's been a Tik TOK meme.

38:22I don't know if you've seen it of, um, putting, uh, various historical figures in the NBA and talking about how like their career is sort of ending. So it's like Queen Elizabeth, like in Celtics Jersey being like, I'm just sort of old now. And like her, like at the podium, I'm talking about retiring and then it cuts to, it's really fucking funny. I laugh every time. Like it's, there's one with a queen Elizabeth, Winston Churchill. Um, uh, who else has there been? Anyway, but so I watched them and I was like trying to show Rosa them and she was like, this is stupid. It's not showing this to me.

38:55But I and I feel kind of gross after I watch them, too. Like I laugh at it. And then in the moment I was like, this truly is slop. I mean, this is the lowest form of entertainment. And I understand these things will evolve over time in terms of acceptance culturally. But as it stands now, like, yeah, it does entertain me. I do laugh, but I don't feel good about it. But the dream is that this is like some Tom specific video, like the dead. They've figured out some sliver of human psychology. And for some reason, you think this odd thing is funny. And then they can target us all and we get all these weird like Freudian like, oh, I like, you know, you have this weird.

39:30I don't know. Yeah. It's just funny to me that you find it amusing. You haven't found many other people who find it funny. It's like this is this is the real dream of these AI generated videos that they can micro target everybody and create the panners to each of us. And it's engineered to be addictive, which is why there's this existential fear in Hollywood over how can we compete against this? If it's all about a battle for attention, it's really hard to say original story versus thing that is almost to the precision of a casino in terms of how it gains your attention. What has been amazing about language models that they haven't gotten enough credit for is that they were all built around the truth.

40:10They were trying to be like, how can we not hallucinate? How can we be as accurate as possible? Unlike social networks, which were built on engagement and said, we'll deliver whatever we can get you to stay engaged. Obviously, AI is more and more involved in that. But we'll use the fact that the content has been created by humans as our excuse that we don't have to worry about the accuracy. Because humans created it, even if we decided to distribute some of it and not others. And now, obviously, something like Sora, you're like, screw it. Let's just do something based on engagement. And that's a dark path.

40:45Well, right. And then the question is, and I talked to Varun Shetty, who's the head of media partnerships and OpenAI about this. Like, are we supposed to view something like Sora as an off ramp as just sort of like some entertainment to the side while you guys are pursuing your path towards AGI? Or is it critical towards getting there? And his argument was the latter. is that like actually you can gain a huge amount of information about the real world through generating videos and uh i don't know the training process that it takes to do that which is controversial on its own um and so that point it's like we'll then prove it like you guys have made this viral sora app that has gotten a ton of people addicted to it it's still number one on the app store like right now open ai has the number one and number two slots on the app store like you guys should really be getting towards agi now like because this is all just an off-ramp towards slop and entertainment and replacing of social media, at the very least, that's disappointing to investors and society at large.

41:45I'm sure it takes a huge amount of GPUs to get this stuff done. But yeah, it's like, I think they need to keep reminding people. And I think it's good for people like Karpathy to keep talking about that. It's like, we know what the goal has to be. Let's keep in mind what the timeline is and be clear-eyed about overhyping it. but like this is not about creating slop i i wanted to talk about you know the new york times did this story on amazon that basically said you know amazon is set on automating a bunch of its workers you know amazon i think is the second largest employer in the united states a company that really believes in robotics already has many robots and they're driving to have even more robots doing work and this is part of their grand strategy and obviously there are robots that are not built on language models but you know the improvements with foundation models and language models have improved computer vision and now in some cases are powering robots entirely and so we're seeing improvements in robots relate powered in some cases by uh ai that uh could mean more automation i'm curious uh what reaction you guys had to that story i mean this has sort of been a long time coming for Amazon.

43:03Like they have been building up their workforce. It's always been this kind of flex situation where they like overhired during the pandemic and then they had to lay a bunch of people off and that caused a huge amount of turmoil there. But there's no question that long-term this business, once technology allowed, you know, robots to be able to put stuff on and off shelves or whatever, that's where it was going to go. So I was kind of surprised to see people mad about it. I understand like, yeah, people losing their jobs is an upsetting prospect. Well, certainly given that it's, you know, to your point, probably one of the largest employers in the US.

43:38But I guess I agree with you, Tom, I was kind of surprised to see people kind of freak out about this story in that it seems like that's definitely was the goal the whole time. I don't know. It's as if it was something new, you know, as if it was something new that automation is where this was going it's just smarter automation frankly for a lot i just hated how ominous the story was like i don't yeah i mean amazon's going to try and do things as cost effectively as possible what what else sort of you yeah i mean you know and this one is more realistic than like drone delivery or other things that you know amazon was more public about but it's weird to see you know i remember like height of the pandemic everyone was looking at the way that Amazon was hiring all these people.

44:25And I remember that movie Nomadland came out where you had Francis McDormand's character working temporarily as like an Amazon, which was done with partnership with Amazon. So it wasn't an anti Amazon thing, but we were all kind of wondering like, Oh my God, is like that going to be the new blue collar job for everyone across the country? And I, it's weird to see people looking at a job that previously we thought was like beneath the dignity of most human beings and you know they're anti-union and that really pissed off a lot of people the working conditions are really terrible people were like you know all the exposés about people peeing in bottles to get the delivery quotas met that kind of stuff a lot of outrage about the jobs before yeah and you know amazon did respond to that stuff and start paying people more you know out of public pressure from people from like bernie sanders and remember aoc was successfully got amazon to not build hq2 in and around new york which kind of was an early win for her in the left-wing movement.

45:21So I, I don't know, like this is just the direction of technology and has been for like dozens of years. They've automated huge portions of assembly lines in, in, in car manufacturing and stuff too. I mean, this is not like skilled labor in most respects. This is putting stuff in boxes and I'd love there to be a world where people that had those jobs had more intellectually inspiring things to do, or there's a UBI that gets people to do other things. I mean, that to me seems a more reasonable conversation than worrying about automation of a job. Companies should automate boring, dumb stuff nobody wants to do.

45:59And we should elect politicians who create a safety net so that every American has healthcare and nobody starves. But we shouldn't handicap great American companies and say they can't use current technology to, you know, get me my, I don't know, battery packs here slightly faster or whatever I ordered from amazon you know in the last last day or two yeah i don't know and true leftists i think understand that as well i mean do they i don't know i just feel like what annoys me about this kind of story is the you know the haters from both sides the haters who are like ai sucks and ai is killing all the jobs it's like which is it that's that's what drives me crazy this sort of like just ominous people who feel like they're driven by like ominous vibes and just like depressives reacting to everything with that as their sort of lens and then find a way to be negative about everything that that's what drives me crazy just socially and interpersonally i realize it's a small concern but that's why these i don't know the the new york times story just had that sort of like negative feel it's like what what yeah yeah i understand the concern and fear about it i mean if we are talking i mean there are cities that were basically destroyed economically because of, I mean, globalization, but also the automation of a lot of jobs like Detroit, in many ways, has not really recovered from many of these kind of third party assembly line businesses, not, you know, meeting as many people on the line as they used to.

47:28And, you know, I don't think the job of a union should primarily be to stop the progress of technology. I'm seeing a lot of this happening in Hollywood too, right? With AI, it's like, what role does a union have to play when you start to see a potential job displacing technology come about. And really your job is to protect the livelihoods of the people who, you know, are affected by it. But trying to stop something from existing just doesn't, it doesn't fit with like the march of time. There's just not great success stories of stopping technology from existing. So the worst ones are these attempts to stop like ports from automating, moving like containers around, which is just insane.

48:07um oh like the dock workers unions and such yeah yeah trying to block you know the automation automation of ports which in our entire economy rests on moving these things in and out yeah i feel like they should also be looking more at like the types of entities that are buying ports in america which are like also fairly insane i mean you have a huge amount of middle eastern money essentially buying up and building new ports in the country it's the only people who are investing and that sort of stuff. So I'd be more interested in like following the money there rather than saying like we need to protect, you know.

48:40Man, I don't want to get like the dock workers pissed off at us and like get that. Oh, really? The comment section is full of steven. Oh, no. I'm not. The dock workers union finding our podcast. Oh, no. I'm not a union diehard. I know. And I'm the leftist on this podcast, too. Wait, I'm the former union maiden. Are you? Yeah. Oh, I was maiden was the title. That was my nickname. No, no, I was, I was, I was involved in the business insider union. And, but I feel like the attitude there has always been, you know, much more just, I guess, not wanting your job completely automated, but it wasn't resistant to AI coming into the newsroom at all.

49:19It was just like, how do we use this strategically? Do we use this to actually write copy? That's kind of the line that we land on, but, you know, for brainstorming. And I mean, I feel like as an editorial rule, you know, I think it's kind of where it's landed in a lot of other news companies to, you know, not having, you know, plug your story into chat GPT and just publish it, you know, not being what's expected of you. Well, within journalism to like the types of jobs that an AI could do or would be doing are, again, not very good jobs. I mean, certainly copy editing. I don't even think those people were part of the BI union, but it's like if they were, but your article are human copywriters.

49:55still generally outperforms Chachi Biti. Chachi Biti, yeah, doesn't catch all the pieces as a copywriter does. It's hard to teach it, obviously, our house style and things like that, but I'm sure that's solvable. Anyway, that's our episode. Thanks, Eric, for dropping on in. I don't know, what are you going to be in for a couple more episodes? What's the plan? I think Cerebral Valley is going to take over the podcast. Max Child and James Willsterman will be back on. We love doing our draft of AI startups. So if you thought this was somewhat pro AI, we're about to take AI by the fire hose for the next couple episodes.

50:38So that'll be a lot of fun. I mean, we're going to be prepping for all these interviews with the CEOs of top AI companies. We have execs from Anthropic, XAI, Replit, Vercel speaking. and so we're gonna be I'm sorry I didn't know this Guillermo is speaking we got some questions for you interesting photo ops that I'm sure will come up at some point in the conversation well we have both Amjad at Replit and Guillermo at Versal speaking and they've feuded a little on Twitter over international politics that is not the theme of our conference we're talking about AI I don't think they're on the same panel no No, no.

51:20And I don't think we're putting them next to each other. This isn't Dario and David Sachs showing up at the Benioff party last week. Oh, how was that? Did they interact? Nope. Nope. And Dario, I overheard him. Here's a blind item for you. Dario kind of muttering to, I believe, his PR person, why am I here? Oh, my God. We love a good pairing here in tech, but specifically at the newcomer Subrepo Valley Conference. Yeah. so that'll be fun and uh this comes out oh yeah this will come out this week i i should have a post in in the newsletter about uh my child i i i've sort of you know there's an instagram post you know they're breadcrumbs but i will talk about having a child and i've quietly been on paternity leave i appreciate you guys uh stewarding the podcast in my absence it feels you had some good episodes and i think we keep learning and improving the show and And we've been punching up production and so excited to pick up from where you guys left off and keep growing the audience and improving the show.

52:25And we love your feedback. We subscribe to the newsletter, follow along, and definitely tell us what's working. We want feedback. What do you like? What don't you like? Send us an email. All right. Thanks, everyone. See you next week. All right. Thank you. See you. Thank you for tuning in to this week's episode of the podcast. If you're new here, please like and subscribe. It really helps out the channel. Listen in for new episodes every week, wherever you get your podcasts.