Gruber Slams Apple, Tech Stocks In Freefall, Vibecoding & OpenAI's Agents

27:16It's really hard to undo free trade. I think we probably could have been more careful in terms of the way we entered these trade agreements. But free trade, as it happens, does benefit big tech in a big way, whether that's advertisers from other countries trying to sell into the U.S. market and running ads on Facebook or expanding in the US and then buying cloud compute here and in particular buying cars. And then you look at Tesla and you're starting to see some ripple effect here. This is from Business Insider. Tesla's latest decline could be one for the history books. It says Tesla has lost so much value in a short period of time that JP Morgan analysts said they couldn't think of another comparable moment in automotive history.

28:01We struggle to think of anything analogous in the history of the automotive industry in which a brand has lost so much value so quickly. The JP Morgan analyst wrote in a note on Wednesday that the historical cases before were confined to a single market, whereas the decline in Tesla sales in 2025 is not specific to any One Nation or geography. So I guess you have the threat of tariffs, which Tesla has now started to say could impact them, and maybe also some blowback to Elon Musk being so close to the Trump administration and associated with his policies that you see in places like, I think, Germany that sales have dropped something like 75 % on Teslas.

28:46It's ironic, I think, that the Germans don't want to buy Teslas. I mean, you know, and there used to be the whole meme about German cars, right? Like if, you know, within the Jewish community. I mean, I think it's sort of, well, Tesla is a special case, right? I mean, it's hard to like separate all of these things out. I mean, people don't like, you know, they don't like Elon Musk. There's all these protests. People are throwing Molotov cocktails at the, you know, at the dealerships. Obviously, a lot of people in the press don't like him either. They're writing nasty stories. That's probably having a ripple effect in the market.

29:24I mean, they've also seen sales drop, I think, in part because they have this whole new lineup of cars coming out, right? There's a slump in January anyway. And then on top of that, their cars are about to be refreshed. It's like the new iPhone is about to come out. Why would you buy an iPhone? So I think there's a lot of things. The company, though, is still, I mean, they're just way ahead of the industry technologically. I mean, the full self-driving technology is just like, it's incredible. Like somebody gave me a ride home from the airport yesterday in a Tesla. They did not touch, all the way from the San Francisco airport to Marin County, they didn't touch the steering wheel once.

30:07I mean, the stuff is, it's incredible. It's running on like transformer models, right? A single transformer model. um so when if tesla starts if they start updating their cars and they're selling cars that allow you to you know not have to pay attention at all like not even look at the road i mean i i think i think that company if people are still going to buy that product no matter who's who's running it i think well let me then put this to you is this a moment and of course this is not investment advice, right? This is just two reporters kind of riffing. So don't buy based off of this advice.

30:45But is this a moment that you'd want to invest in companies like Tesla as they're being hit by sentiment and maybe the Magnificent Seven as they're sort of, they're being put in the middle of a potential trade war? Now, I think the optimistic view would be that a lot of this is just posturing or at the very worst, it's Trump trying to get some of the tough stuff out of the way in the beginning of his term. So he looks like someone who's really saved the economy towards the end and left it in good shape. Do what you have to basically crush inflation and then build from there. And in fact, I asked you, Reed, if there are any stories you want to discuss today.

31:25And you said, it's a great time to invest in AI stocks, thanks to Trump tanking the market. So I turn that to you. Is this the moment? Right. I mean, yeah, back to not get, I mean, look, I don't, I don't invest in individual stocks. I basically never have because, you know, being a business reporter all these years doesn't allow me to, but it does look, you're just Tesla aside. I mean, I would say Tesla is probably an AI stock, but Tesla aside, like, you know, there, there, there's huge demand for this new technology. I mean, at the end of the day, something very drastic would have to happen to change the current situation, which is you have a new wave of technology that is going to completely upend the world economically in the long run.

32:19It's just a question of how fast it will happen. in all these companies that are, you know, building the infrastructure around that are, they're kind of no brainer investments in my mind. But it's easy for me to say, right? I'm not, I'm not like, you know, putting my money, I mean, I have whatever, a 401k, but how, how does that not, like, how does a company spending a lot of money to build a new technology that has tremendous demand? Like, how does that not make sense? You tell me, I, I can't, that you know no i think it does make sense it makes sense as long as it delivers and i think we're still in this point where we're not exactly sure the magnitude that ai will deliver okay i think we've gotten over the hump of will ai be useful oh it sure is yeah but we don't yet know like let's say what percentage of gdp growth ai will contribute to and if it's not a meaningful percent, then a lot of this investment will be like, all right, well, you know, you can upload your tax returns into chat GPT and have a fun conversation with it, but it's not going to like, you know, help a company make 10 % more revenue.

33:31Well, wait, there's two things. There's two things there that I think you're, I think you're combining, which is like GDP growth and companies making money. Like you don't, you don't have to, to make a lot of money. You don't have to move the GDP of a country. No, without a doubt. I was using that as an example to illustrate the cumulative productivity increases that we're seeing from AI. And so the argument for those who support AI is that it will be profound enough that it will lead to tangible nationwide productivity increases. Of course, a company can make money that you can make money again off of these like, pay ten dollars and you can put jd vance's face on a lemon but the question is the magnitude of this and that's that to me is like when you ask like can should we question the wisdom of this technology that's going to change everything it's like the reasonable doubt that you would have here uh is is not that anything that's been done politically will change what's been happened where the trajectory is, but just like how much they live up to the big promises that some of the AI labs have put out there, like Dario Amode saying, you know, there's not going to be a software engineer in like three months.

34:44If that's the case, that's a magnet, you know, very different than if it's, again, just this thing that you can kind of joke around with and, you know, tell your secrets to, and then it can give you some advice. Right. I mean, what I'm saying is I actually don't know what the GDP impact is going to be of AI. Because the thing that people always forget is you can invent something that's amazing, but people have to adopt that technology. And generally, we're very slow to adopt technology. I mean, if you look at the internet or just the personal computer internet web sort of era of the last 35 years, I mean, it's a massive change.

35:26It's completely disrupted the job market, right? But it happened gradually. It wasn't like it just smacked everyone over the face and instantly all these jobs ceased to exist. But it did change. There are jobs that don't exist. A lot of jobs don't exist because of that new technology. And I think AI might happen faster. There's different scenarios, right? If all of a sudden it just happens so fast and you just have no choice but to adopt it and And, you know, that is probably not a great thing, honestly. That might really affect things in a negative way because I don't think humanity is really great at sort of, you know, adjusting to these really fast changes.

36:10But, you know, over time, for sure, right? But then at the same time, like, you don't have to have – like, if even just software development gets totally disrupted, if that's the only thing that comes out of this, which seems like pretty likely at this point, like that's enough for these companies to make a ton of money. It's still like a good investment, I think, like regardless of what the big long-term impact is on the world. Does that make sense or do you disagree? I only disagree slightly. I think that the money that we're seeing being put into these companies would be very different if the use case was only software development.

36:57Are you still going to make a company that is extremely profitable, potentially? But are you going to make something that's going to live up to the expectations of investors if it can just do software engineering? I don't think so. And I don't think that's what's being pitched to the investors either. No, no. I mean, you're right. And I think it is much more than that. But even that is like, you know. I'm not going to deny that's big. Yeah. Yeah. I mean, like if anyone can hire, if anybody can have an army of software developers for basically nothing, like that's pretty, it's pretty impactful.

37:31I agree. Having been as deep in the technology as I have been, I'm sure you've done the same, trying it out, testing the new features. It is incredible what it's doing, like how quickly it's moving and how good it is already. but again there's one more question overhanging all this which is how much better is it going to get because those concerns about the you know the diminishing returns on pre-training and growing the size of your models those are legit jan lecun's going to be on the show next week we're going to talk about it um he's pretty convinced that we've sort of we have hit that moment of diminishing returns and hearing him speak to it, it's like, okay, so we're going to need some more breakthroughs to keep having this technology be more useful.

38:23And we don't know what those breakthroughs are. So there's a chance that we just kind of hit a cap of what LLMs can do today. And of course, then you go to productizing them. And that's the big debate we have on the show all the time. It's the product or the model that matters. But I would say it may be trouble, again, if this is the best they are. I'm not complaining. They're very good, but they need to get better to, again, live up to these expectations. Yeah, there's totally a leap of faith thing that happens. And I think Jan is so smart and love his, just anything he puts out there is fascinating to listen to.

39:01But I think that And there's this leap of faith thing that happens with AI where you build more compute capacity, right? You allow AI researchers, computer scientists access to just more compute power, and they just come up with new ways to use it. And so, yeah, I mean, of course, if you ramp up, if you just increase the compute power by orders of magnitude overnight, I think it looks like there's a plateau because it takes time. Like normally that happens more gradually. So like the AI research and the compute power kind of go together over time. Like they, you know, it's like a ping pong. There's a little more compute, a little more research, a little more.

39:48Right now there's just this huge leap in compute. So I think the research hasn't quite caught up to it yet. But again, like that's not something you can prove. Like you can't prove that there will be some breakthrough in the future that some kid who's graduating from, you know, Stanford today is going to figure something out two years from now. But I do think it will probably happen. And I guess we're going to talk more, you know, about some of the new technology later, too. So I don't want to get like too much into that now. Well, that is a great tease. So we are going to talk about some of the new technology, including the development of proprietary chips from Amazon and what that's going to lead to, how that's going to advance AI.

40:29And also, we haven't yet talked about how OpenAI is going to let other apps deploy its agent AI. So we'll do that when we come back from the break right after this. Did you know your credit card points and miles can lose value to inflation? Credit card companies often reduce the redemption value of your points and miles. Now, imagine a credit card with rewards that can grow in value. With the Gemini credit card, you can earn Bitcoin or one of over 50 other cryptos instantly with no annual fee. Every swipe at the store or gas pump earns you instant rewards deposited straight to your account. Plus, sign up now for a$200 Bitcoin bonus to kickstart your rewards.

41:11Visit Gemini.com slash card today. Check out the link in the description for more information on rates. Again, if you're looking to invest in Bitcoin but don't know where to start, the Gemini credit card makes it easy. The Gemini credit card is issued by WebBank. In order to qualify for the$200 crypto intro bonus, you must spend$3 ,000 in your first 90 days. Some exclusions apply to instant rewards in which rewards are deposited when the transaction posts. This content is not investment advice and trading crypto involves risk. The Gemini credit card cannot be used to make gambling-related purchases.

41:49You're used to hearing my voice on the world bringing you interviews from around the globe. And you hear me reporting environment and climate news. I'm Carolyn Buehler. And I'm Marco Werman. We're now with you hosting The World Together. More global journalism with a fresh new sound. Listen to The World on your local public radio station and wherever you find your podcasts.

42:16And we're back here on Big Technology Podcast with Reid Albergati, the technology editor at Semaphore. We have big news coming out from OpenAI. This is from The Verge. OpenAI will let other apps deploy its computer operating AI. Agents are said to be the future of AI, and now OpenAI is trying to help developers build their own. The company is releasing a new responses API that offers building blocks for developers to create agents capable of searching the web. digging through files and performing tasks on a computer on their behalf. There are some agents that we will be able to build ourselves, like Deep Research and Operator, says Oliver Godeme, the head of product for the OpenAI platform.

42:57But the world is so complex that there are so many industries and use cases, and we're super excited to provide these foundations, building blocks for developers to build the best agents for their use cases and their needs. I want to say that this is big because giving this technology to developers, to me, seems like a real opportunity to be able to sort of prove out the agent use case we've been hearing about so often. But the skeptical side of me says operator and deep research seem more like a party trick than an actual application right now. and I guess it might be useful in some maybe obscure data entry case, but these companies are already using macros and robotic process automation and I don't know if this will necessarily be a leap ahead.

43:49You've been following the news. What do you think about it? Well, yeah, I mean, I think you bring up a good point. I mean, look, I think it's probably big just because you're giving developers, or is you're giving the general public access to something that will allow them to just unleash their creativity on, you know, on, on this technology. And I think that's always interesting. Like, it's just sort of exciting to, to kind of like see what people come up with. I haven't played around with it yet, but I don't know if you've done this. I, I learned that it's called vibe coding. I didn't really know, but like I've, I've indeed.

44:27Yeah. I mean, I, I started doing that and um you know it's just like it's incredible what you can do you know with you know when just playing around with this stuff is somebody who just doesn't really know anything about coding what you can build is it's amazing but it's also you can kind of see how like again back to like the infrastructure build outs you could see how you know you start hitting these token limits, these context limits that sort of, you know, I think limit the technology right now, but like, well, those barriers will, will be, you know, as long as people keep investing in the, you know, in the infrastructure, which I think they will, those barriers are going to go away.

45:10And then I think that's when it becomes really, you know, even more transformative. Have you vibe coded anything fun? Yeah. Like I did some stuff with, uh, Which we define vibe coding. It's basically just prompting software and just adjusting based on prompts, not coding things and just building anything that you want in your daily life. Right. In my case, it's just like, it's like, okay, I'm going to try this. Like, let me see if I can build a little app with Replit, you know, Replit agent. And then eventually, you know, I need more tokens. I need more power. You know, I'm going to go use, you know, Claude Sonnet.

45:48And it's like, you know, create software that does this and you watch it like write code. And then, you know, you're putting that into terminal and you're at the same time. I mean, for me, as you know, somebody who's covered technology, but never been a technologist, it's like I'm learning so much, too. It just it teaches you like in a way like how software works if you don't know already. and I just find it if you do that and you start really getting into it and you start hitting these limits of the token limits and stuff and you're getting cut off it's like oh okay I can see I kind of see now what this is all about and why we need all these data centers and you know to bring the cost down but and so like I just when I see that I'm just like oh I want to try that Like, I want to see how that works.

46:44What have I built? I mean, you know, the thing is, it's not there yet. So these things aren't, like, super complex. But, like, just stuff to help me, you know, help me plan kids' summer camps, right? Like, going out on the web and pulling, you know, pulling information off of various, you know, websites and, you know, putting it into different forms. I thought the one fun was just like looking at like having operator look for specific emails and then like connecting that to Twilio and having it send me a text message about it. I mean, you could just do cool stuff. And it's not, but again, like I want to build more complex things, right?

47:32I want to like find ways to save lots of time and automate everything from my professional life to my personal life. And we're not quite there yet, but it seems kind of close. Have you built anything? What have you done? Yeah, I mean, I've built, I mean, using Claude, and it's just a couple of prompts, but I've built games, like I talked about on our Wednesday episode with the Roblox CEO. I also built a retirement, like a retirement calculator where I uploaded a fake bank statement and was just like, make me a financial plan based off of this bank statement. And I was like, all right, well now build me a retirement calculator just for kicks and it built a working retirement calculator uh with like various fields like a very nice and sophisticated one so to me i think that like just this power to imagine and prompt any piece of software like as we're talking i'm thinking like do i want to build uh like my own tracker for big technology stories or podcast guests a specific calendar like you can probably prompt that with claude and they give you dedicated links so you can kind of use that uh as a web page and sort of next thing you know you have this like custom application that you would have had to pay maybe a few thousand dollars for someone to build for you yeah totally and i think it's um have you have you played around with like any of these ones the like like the replit agent there's a bunch of them where they'll like literally just built you just like build me this and the app is just created yeah i mean that's what Claude can do that now you did that with Claude like totally end to end yeah and to end Claude yeah that's a that's that's really fun yeah I mean Replit uses Claude so I guess it's similar but yeah yeah no it's it's crazy and and like when we talk about how things are going to get better uh that in particular to me is is very exciting and and it might be lucrative so it's like again talking about the payoff now we have a story it's a little bit old but I think we had it's worth talking about on the show that open AI is reportedly planning to charge up to$20 ,000 a month for specialized AI agents.

49:32So they're working on a high income knowledge worker agent, which will be priced at 2000 a month, a software developer agent at 10 ,000 a month and the most expensive rumored agent, $20 ,000 per month for a PhD level research assistant. The story, I think this is the information story. They say that SoftBank and OpenAI Investor has committed$3 billion to use OpenAI's agent products this year alone. So maybe this is an answer to our question from what's going to happen from the first half. Do you think this stuff is going to get any uptake, Reed? You know, the$20 ,000 thing, it hasn't actually launched yet.

50:17But my guess is if it does launch, and I did read that story and the information, great great reporting as always um if it does if it does launch then i think they know you know there probably is demand for it right and i mean it's amazing that i mean there's they have like the 200 a month plan too right the pro plan that oh yeah apparently is just you know has incredible uptake and they're still you know they're still losing money on it to be fair but i mean the fact that people are willing to pay that much for this stuff is just, you know, and it's, and it's still so new. I mean, it just feels like, it still feels like V1 to me.

50:56I mean, when I, when I play with it, so, you know, it's just, it's, you know, it's an amazing, it's amazing technology, I think. Definitely is. Have you tried or do you have any thoughts on Manus? I think it's called Manus, Manus. It's the Chinese agent. You know, I haven't used it actually. I tried, I tried deep seek, played around with that. But I haven't, have you used that? I haven't used it. I've just seen it working at like incredibly fast paced speed. And I actually read this, I mean, I've applied to be on the wait list, but there is apparently a lengthy wait list. And I don't know, I've been, I've been too busy this week to write them and say, hey, can I get like a press access point?

51:41But MIT Tech Review did get access and I read their review. And they found that it feels like collaborating with a highly intelligent and efficient intern. While it occasionally lacks understanding of what it's being asked to do, makes incorrect assumptions or cuts corners to expedite tasks, it explains its reasoning clearly. It's remarkably adaptable and it can improve substantially when provided with detailed instructions or feedback. It's promising, but not perfect. Okay, so this journalist gave Manis three assignments. One of them, let's just talk about one, compile a list of notable reporters covering China tech.

52:18So this is the review of how it did. It gave them a list with five names and five honorable mentions below them. Some of the journalist's notable work was listed, Some of it wasn't. The journalist asked Manus why, and it said it was lazy. Partly due to time constraints, I tried to expedite the research process. This is my, again, this is my criticism of this stuff. Yeah. You could probably just ask ChatGPT for a list, and it will do just as good a job as your fancy agent. So I think we're going to just have some moments where we try to have to figure out how to use these. And like you said, it's going to take a while for the use cases to develop.

52:58Yeah, and I think the other thing that's sort of about the Chinese ones, like DeepSync, I mean, DeepSync had this app as well, is that, you know, even OpenAI, like, do you remember Sam Altman tweeting, like, we're sorry, we're out of GPUs? I mean, it's like, I mean, I'm sure you talked about it on the show. And it's like, you know, they're blocking GPUs from being sent to China. So imagine those, you know, if any of those products took off, and they would just have a very difficult time, you know, keeping up with the demand. So it's all about efficiency, right? Like, it's all about creating, you know, doing this stuff with as efficiently as possible, which is great but like you know even the state of the art stuff like Claude Sonnet and etc is you know pushing the the barely you know it's still like v1 right so if you're doing the sort of like slightly less than you know the state of the art but more efficiently it's like it's never going to be like that impressive right because it's always it's always kind of like a slight step behind.

54:05So, you know, I, that's sort of how I view these products. Like they're, they have to get more efficient. Everybody's trying to do that. Everybody's trying to make this stuff as inexpensive as possible. Um, the, the, the big labs are doing the same thing. Right. So I don't know. I mean, I think they're, they're interesting, but it's not like, it's not going to like blow our, you know, blow up the whole market, you know? Yeah. And it's just going to take a while for use cases. I mean, I think criminals will have a great time with this stuff because it will just try to fish everybody and hack into their systems.

54:42But the good use cases, I think, will be tougher to find. So right before we leave, I want to hear you have a story coming out in Semaphore. It's about Amazon's chip effort called Project Rainier. And I've always wondered about these proprietary chip efforts because you hear the same thing from every CEO that talks about them, which is we'll take every chip that NVIDIA will give us and we're very confident in our own chips, which to me is somewhat contradictory. So what have you found and which part of the executive statement should we believe the most? Well, I went out and visited the Amazon chip lab and talked to them.

55:21And obviously, they want to talk up their technology. The interesting thing I think about, there's a couple of interesting things about the Amazon effort. So they've been at it for a while now. They acquired this Israeli company, Annapurna Labs, 10 years ago. and this is their second generation of the Tranium chips, Tranium 2. And this chip is, I mean, it's not as good as like an NVIDIA chip on a one-to-one basis if you compare them. But if you network them together and you're talking about hundreds of thousands of these things and every piece of the network of the data centers are custom built to be as efficient as possible, you do start to get sort of these, these efficiency gains.

56:11And that's kind of what it's all about, right? We've talked about this now, you know, this has come up on this show a few times. So this, their big project is called Rainier, right? This, they're going to build this huge cluster of chips. And they say that it's also, they've found a way to connect multiple buildings together, you know, using, using this elastic fabric technology that they've been at for a while. If that's true, I think it's really interesting. I mean, to my knowledge, people have trained models across multiple locations for sure, but the state-of-the-art models, either they have to be split apart where parts of the model are being trained in one building.

56:56But what Annapurna is saying is you can actually treat a multiple location cluster of compute as one massive cluster, as if it's in the same building. I mean, if that's true, then I think that puts them a step ahead. And then the other big thing is, and this is not new, but they've got Anthropic to train the next model of Claude on this Rainier cluster. And it worked. What's that? It worked. well they know they haven't they haven't actually built the the cluster hasn't even built yet so i think this is like it's like a big test i think for for amazon if they can actually pull this off and the other thing is like they're a huge investor in anthropic so you know it's not like a i mean i'm sure there's you know that has something to do with this decision but i also don't think Anthropic is just going to like, they're not going to like decide to, you know, to stake their future of the company on this technology if it's not good.

58:06And so I think what happens, like what could happen with this effort is that, you know, and also Anthropic has helped them design these chips and all that. And what could happen is, you know, if this is a success, if they're able to build Claude and Claude remains, you know, state of the art, that gives them a lot more customers, right? Because now Claude will run more efficiently on the Tranium 2 chips. And so I think that that actually does start to help. And they've got some other, you know, some other customers as well, like Databricks, Apple's considering it, you know, Poolside, which is like a kind of hot AI company.

58:46So I think there's an interesting, I mean, they definitely have an interesting story around these data centers. And they're not the only ones. They're not the only ones out there. I mean, Google is obviously, they've been doing this the longest, this kind of vertically integrated custom chip strategy. Meta's getting into it reportedly. So there are other people doing it. But I mean, if you think about this, the scale of this stuff, efficiency does become kind of the most important thing. so I do think maybe not every kind of every custom chip out there but the ones that where you can really vertically integrate are really interesting and so what happens to NVIDIA well NVIDIA is fine I mean I don't think the pie is growing so big so it's growing so fast like nobody's going to like lose here you know none of these big players are going to lose like NVIDIA you know people are still going to like NVIDIA still won't be able to keep up with demand for their chips I think for, for a long time.

59:49I mean, it's just that you need more. Like it's just not, you know, you can't meet demand with just NVIDIA, I think. Definitely. All right. We will set this podcast to go live just as your story comes out. Tell folks where they can find it. They can find it on semaphore.com and please sign up for my newsletter. It's free for now. And are you guys going paid? I just I just like to I don't want to make it I want to create some urgency maybe you'll get grandfathered in when we do I'm subscribed I get it every time it gets sent out and I do suggest you go check it out as well Reid Albergatti great to see you again our returning champ delivers another championship performance here thank you for coming on the show thank you so much it was really fun super fun alright everybody as I mentioned Jan Lacuna is going to be on the show on Wednesday.

1:00:45So stay tuned for that. You're not going to want to miss it. And then next week, Ronja and I will be back to break down the week's news. Thank you for listening. And we'll see you next time on Big Technology Podcast.