Swyx on Meta's AI talent raid: $100M+ first-year packages, OpenAI's surprising resilience, and Cursor poaching Claude Code's creators

VCs are the new whalers. Yes. Yes. They really are. And the old Whalers. And the old Whalers. That's true. Zach, this was so much fun. We'll have to have you call in tomorrow or Thursday. Yeah, when there's more news. Please. I'll be around. It's my first wedding anniversary, so I'll be around.

Plenty of time to call in. Thank you very much. Incredible. Incredible. We'll talk to you soon. Thanks for having me, boys. See you soon. Yeah. Bye. Up next, we're going back to the AI talent wars, talking trade deals, talking$und00 million offers with Sean Swix. How are you doing? There he is.

with the suit looking fantastic. We You wore a suit last time and we didn't give you enough credit because you were at your own conference uh the AI World Fair and I was just This is a sign of great respect. This is a great respect. So, thank you. Thank you. And it's always great to great to catch up with you.

Um how how did you process the news over the weekend? What jumped out at you? I saw a couple of your posts. You were breaking it down. Um but but uh h how would you tell the story? How how how big is this for Meta? How bad is this for OpenAI?

Um how are you feeling about the trade deals that we've seen break loose over the last few days? I feel like people who have been in discussing like sports trades uh might be way more equipped to to discuss this.

this is like relatively new in in AI land and like honestly just like absurd uh you know like no like get that back guys like you know like I'm I'm happy you're getting paid but like also uh it definitely feels like for uh the a lot of the AI researchers I talked to like like this is a little bit um sort of like a panic move and and in some extent warranted right like I think that Meta has um felt like it needs to do something different than what it has been doing especially leading up to Llama 4.

Uh and this is different. This is very very different. Uh this is something that only Zuck can do I think. So there was just just for some extra context there was new reporting from uh Kylie over at Wire just now saying that uh you know the there's been some back and forth. What are the actual offers?

Obviously a lot of them are different, right? It's not just like a set offer that everybody is getting. If you join Meta, you get $100 million immediately. In a briefcase, hand delivered.

I don't think that's what's so apparently some of them are up to 300 million over four years with more than a hundred million in total compensation for the first year. This is Wired's reporting, right? Mark Kerman was saying more like 10. She said she needed to see an offer letter.

But the funny thing here is there's an OpenAI staffer who was responding to this anonymously and they said that's about how much it would take for me to go work at Meta. shots shots fired. Wow. But I mean they have set right so now they have unfortunately they have set their own price. It is public now.

Um I think there was some back and forth between like whether or not Sam appearing on his brother's podcast was like a very calculated move and like was he dishonest? Uh this is not my words. Is this what it seems like he might have been honest based on this reporting. Like it seems like it's real.

I think it look in that ballpark, right? like whether or not it's exactly 100 plus minus like it is at least eight figures and that is like far and far and beyond uh what any anyone else had been getting prior basically and it it is in there at least in the relative range of what Tim Cook and Satia made all of 2024.

Yeah. Which is which is crazy. Also with the 100 million number, it's so round that it just jumps out to you. And I think it's much more viral than 112 or 89. Like 100 is just such a round number. It's such a big amount of money. It's just instant generational wealth. Uh for sure.

Uh I think like that's exactly um it's weird that the calculus of like does this make sense for meta? Yes, it does because uh I think uh you guys covered my uh analysis before that like basically they just cut VR spend and reallocate to AI spend and they have the funding.

They have the equivalent of anthropics entire annual spend. Yeah. Meta's at all-time highs right now, right? The the market and so and so it doesn't really m if if you could move a $1. 8 trillion company market cap by a couple points, you're doing pretty well. Yeah. Yeah.

Yeah, Zuck is no stranger to betting like 1% of his company, right? To totally against the future. Um, and he like I think of uniquely of of a lot of the sort of fang or what is it? I think a lot of people are trying to come up with like new acronyms for like the the Magnificent 7. So, I like the one that's uh among us.

Among Us. Yeah, Among Us is the best acronym. Uh, there have been some wild acronyms flying around you're talking about.

like Zuck is uniquely empowered as as like a you know founder mode of the of the trillion dollar companies uh to actually do this and like so like uh it's really amazing that the leaders he's hired and the people that he's managed to hire.

Um I I would say that uh you know to to one part of your opening question like how effective is OpenAI? I would have said no. Uh not not that affected. They have like super deep bench. Um even like the people that you don't really hear about like are actually like fantastic leaders in in their own right.

Um and so like in some in some in some extent opening eyes is like very robust to to already sort of leave like departures already. Uh but the email from uh Marchen over the weekend I think was also like oh like I wouldn't I wouldn't say that if I were him. I don't know if you you guys have the the saw of you. Oh yeah.

that knowing that that memo would immediately leak, it felt emotional and defensive versus, "Hey, John John's phrasing was was or or at least the way that you you said if you phrase this as, hey guys, we have billions of people coming to our website monthly. We have hundreds we have billions and billions of revenue.

We are the category leader in the space. We are doing something different than Meta is doing and we will continue to dominate and now is the time to rally. This is the fourth time this has happened. Maybe fifth time this has happened. So we survived Anthropic and Daario leaving. Don't forget surviving Elon Elon leaving.

We sur we survived Ilia leaving and we survived Meera leaving. And there's four spinout direct competitor co-founder companies. This is a couple people leaving. We're going to get through it. We've been Look at the growth chart. We're growing, growing, growing.

I would have Yeah, definitely gone from a source of strength there. I don't know.

But yeah, what is your thought on like the narrative that like um OpenAI seems Lindy at this point incredibly resilient, able to lose key people and just keep plotting forward with, you know, iterative releases, models get a little bit better, UI on the app gets a little bit better, deeper research comes out.

They're just always staying on the frontier, just one step ahead of everyone else in that core chat product. And yeah, maybe they maybe they might not be able to dominate in, you know, the the clawed code ecosystem immediately. They got to make some plays to make that happen.

They definitely want to, by the way, they want to. Yeah, for sure.

But in terms of the core thing that's clearly dominant and huge and like gamechanging and profitable like they're doing great and they should and they should kind of uh they should lead from a source of strength there u because they've been through so much and continue to deliver on that front and and and this doesn't seem like a cause for like the narrative around this this leaving is that okay Meta is going to catch up to a frontier LLM a reasoning model.

They're going to have great image generation, but does anyone think that, you know, in three years time people's f people's home screens are going to switch from chat GPT to Meta Chat? No, I don't think so. But what about you? Yeah. Yeah. I I don't think so.

Uh keep in mind also Meta has its own institutional issues with AI products. Uh I think they launched this like um AI chat feature that where every chat was public. Uh I don't think I still think it's still there. Well, I think it was accidentally sharing a box you had to check.

Yeah, people were accidentally sharing it. But it kind of speaks to like yeah, when you're dealing with trillion DAUs, like you're going to have weird stuff happened.

But yeah, uh I would say that Sam would be the first uh of many people to acknowledge that there's such a thing as a 10,000x uh engineer and 10,000x AI researcher. Uh but they they also have a very deep bench and it's not just about the bench, it's also about the data.

It's also about the infrastructure that they've accumulated um that you know the TM um what's their what's their acronym um MSL um yeah meta super intelligence is going to have to build up effectively from scratch because if effectively Yanukun has uh failed in his in his leadership of of uh meta AI there um and I think like that that's fine like that's that's a pivot like that's that's something that only Zuck can do but like yeah open is a very deep bench and like I think they are um like they're they're they're good there.

I would say though that um it's a really interesting like how well diversified this this uh landscape is. Most people when I talk to uh you know there's there's often uh you know in AI circles there this discussion of who has the mandate of heaven.

Mandate of heaven meaning they're just like most likely to achieve a obviously uh of the big labs myself excluded uh of the big labs a lot of people are mostly saying deep mind uh is actually not open. So I think like uh OBI is like in intensely aware that they're being um you know poached.

They're being out competed on on on pricing. Um they're being out competed on some metrics like uh elements for deep mind. Uh and they're and open is like extremely competitive. Like whenever I talk to them, I'm so impressed that like they're not just arrogance. They're not they're not like resting on their laurels.

They're actually just like we will win on every single front. Um and I think it's just a very impressive company that way. Then let's flip over. I mean, MSL is a very it's a cooler acronym than than Llama. I always love Llama because it's a fun name, but it it didn't have the same like seriousness. Fair.

And fair felt also just like it's it's a little nice. It's not hardcore team. No, we don't we don't care about fairness. We We dropped safety from MSL is like going to the majors. Exactly. Sounds like major league LLM, major league AI training.

Um but but I want to talk about the historical energy analogy because Zuck has been in this position before uh with the shift to mobile.

So when when Meta or when Facebook back in the day was a web browser, they initially moved over to uh to an iPhone app and the iPhone was growing and they kind of made a mistake going a little bit too far into the HTML 5 route, wrapping it in an in an iOS shell, but they didn't build the app natively and they were kind of falling behind.

But then, you know, from 2012 to 2015, Mark was extremely aggressive about acquiring Instagram and WhatsApp, rebuilding the team, hiring some great people.

This is when Adam Masseri and andrew Bosworth came in and really really built out monetization in the feed, so they had more money flowing through the system to actually pay for all this stuff.

And then yeah, we were covering yesterday that the the Zuck did a talent raid in the 2010 era of Google because he just wanted to get better at ads basically and then he built the greatest ad engine of all time. Yeah.

Uh so yeah, I mean yeah, how are you how are you interpreting like the comeback story and the fact that like Mark has also been here before? I think that it's interesting how you have like Zuck has almost has to fail first and then he overcorrects and and goes really really hard and wins.

Um in contrast I think like VR never really had a failure because it didn't really uh you know have that had that uh uh that much of a of a success to begin with. But like um yeah like um MSL could only have happened after the comparative disappointment of Llama 4.

Um if Llama 4 had had proportionately the same traction that Llama 3 had, I don't think we'd be sitting here today. Totally.

So actually this is like net positive that we went through that because they were like oh like this like we proved out that this is we've tapped out in this direction and we just need a whole new team uh and we need to scale things in a very different way and probably we just need to poach from the other labs.

what they've done. What do you think the Oh, sorry. Yeah, continue. It's interest It's also interesting though that like what their their current strategy was their third choice.

Um and somewhere in the middle either first or second choice was perplexity and that would have led um that would have led um Facebook or or Meta in a very very different direction more consumery than right now which is building a lab. What do you think the postmortem on Llama 4 is?

Is it just that they over rotated towards pre-training massive model behemoth these like you know not focused on reasoning not focused on all the little tricks and tool use and stuff that actually creates a delightful LLM experience or chat experience these days or is there something else going on there like how do you tell the story of of what they got wrong with Llama 4 if anything I don't want to bag Llama for too much I by the way I put in a a thing in the chat for for people to compare.

Um, Scout and Maverick were great and then the the shenanigans around Ellen Marina were regrettable but like understandable honor at mistake whatever. Um, but uh it seems that Behemoth will just never be released. Interesting.

Uh the assertion again behemoths was never released so this is all rumors uh was that they tried to basically adopt whatever that Deep Seek did for their large uh fine grain uh mixture of experts model and it just didn't work. They didn't they didn't portrain it enough whatever. Um we will never really know the answer.

Uh I think like there's probably also organizational issues around what data they can use. Uh I I know this for a fact just because like I I've heard issues around like honestly like at a big corp when your lawyers get into the mix of training data issues that can get very inconvenient.

Um when uh you know I think like comparatively anthropic when they won their recent case uh they were more like you know seek uh you know bend the laws and seek founder mode founder mode ask for forgiveness.

I think now that we actually have a little bit of ruling around like okay like pre-trading on on copyrighted data is actually is transformative use therefore it's fair use therefore it's fair game go ahead uh actually we'll see but you the case is still trying whether that you need to buy the underlying thing which could still be egregiously you know it could be incredibly expensive to to acquire the copyrighted material not at all you don't think you don't think that'll go through no no whether or not you underlying thing is fine.

It's actually just the the issue of copyright whether you have to pay per use. Uh it's actually pay per inference. Um now that that issue is out of Yeah.

But I I was saying the anthropic case is still like they had 7,000 7 million books that they obtained through that they don't that they didn't purchase and the the pirated ones the pirated ones.

And so it will be interesting if labs have to effectively strike deals with publish you know I don't know it's it's still in the course they they are doing that uh they probably just just should do that. Uh it's a onetime cost comparatively comp compared to the value that they can extract out of the books.

The books are just guaranteed to be underpriced because the books are priced as though they're sold to one person whereas you know a model is going to be used by millions. So like there is there's just there's just no like it's fine just pay it.

Um the the piracy thing is a is a legacy of books three which is a data set that was created in a in by let's say someone who was very pro piracy. Uh uh it was it was a copy of books one and books two which is actual uh licensed uh public domain books.

Anyway, so so long long story short, like I think I think those uh cases will be resolved and like um I don't I I think that Llama and Meta's strategy there were probably affected by the quality of the data that they had um and the fact that they just couldn't they they like sort of tight they were playing one hand tied behind their back whereas all the other labs were not.

Yeah. Yeah. Interesting. Well, this has been fantastic. Last last one one more question. Any reactions to the news that Apple is working with uh you know potentially working on a deal with Anthropic or OpenAI? philanthropic wants potentially billions out of it.

Uh very very convenient timing for the open model release uh which is rumored to be soon. Um uh but like I no reactions uh yet. I think uh Apple there's a lot of news around Apple that never materializes. I think Apple's hasn't shown that it has gotten the memo like Zuck has shown.

Um so until until it wakes up like I don't know. I think like around the time of the Apple intelligence launch last year.

Um I think a lot of people I I was I was commenting that like basically Apple wanted to be the first layer of intelligence and and it wanted to route uh so it always wanted to keep wanted to be Switzerland wanted to play nice between open topic and all the and uh whatever other labs probably not Google but um I think you kind of have to make a bet.

I think you I think you kind of have to go all in. Like I'm I'm not that interested in in Apple making deals with OpenIthropic. I still want open eye to do an open eye iPhone. Uh I don't feel like I don't like the in your device.

Um uh and I and you know I just I just wish they just went all in and and took Apple on head on. Uh because like all of these are half measures, man. Like the like we're going to be sitting here like five years from now and Siri is still going to like kind of suck.

Um like it's the same deal as like Amazon making a deal with Enthropic for like smarter Alexa. Where is that today? Yep. I was just asking. I haven't I haven't actually seen it in the wild. Yeah, great point. This is really one more breaking one more breaking news for you guys.

I don't know you probably guys probably missed it but um cursor has hired the cloud code people from the topic. What do you mean? What do you mean they hired like how many of them? I heard they hired the notion mail team. I didn't realize that they hired the the cloud code team.

That's an that's an interesting commentary on like the like what is is notion able to keep its people because the notion mail team was uh you know they they just launched it. Yeah, they just launched it and now they've left trade deal. But like cursor so cloud code developers have been talking about it non-stop.

I think uh people who are like adjacent to developers really pay attention to this one. This is as close as anything is to the new being the new cursor. It is cloud code. Uh people are