Beyond Content: How AI Is Rewiring Marketing

Transcript generated from YouTube's automatic captions and lightly edited for readability (filler words and false starts removed, names corrected). Speaker labels were added by hand and may contain errors.

Rob May: Hi everyone. I'm Rob May. I'm the co-founder and CEO at Neurometric.

Lauren Slyman: I'm Lauren Slyman. I'm a user researcher at Microsoft.

Anna Kirk: I'm Anna Kirk. I'm sales lead at Thread AI.

Rob May: And you're watching the… in the AI in NYC Show. Well, you'd think as many times as I've said that I would get it right, but you've got to let people know that you're not a deepfake.

Lauren Slyman: Yeah, that's true. Your humanity coming out.

Rob May: Errors prove I'm not an AI. All right. So, hot take to start this episode this week. We are in New York City talking about the virtues of building companies in New York City, and the Oxford 2026 Global Cities study just named New York the number one city in the world, ahead of London and Paris. What do you guys think about that? You guys agree?

Anna Kirk: I don't know about ahead of London.

Lauren Slyman: Yeah, isn't it? It's similar to London, but…

Rob May: I've been there for like eight hours total.

Anna Kirk: I've been there a bunch. Better nightlife than London. Definitely better.

Rob May: Better economic opportunities. So where we ranked — we weren't top on everything, right? Pollution was meh. Crime, we weren't bad, but we weren't the safest or the best. But things to do — clearly this city kicks every other city's ass in the world on that. And then economic opportunities, and then depth of the talent pool, across the board, not just in one area.

Lauren Slyman: Is there like a cockroach index anywhere, or was that weighted?

Rob May: I don't know. I didn't see that, or mice or rats or anything like that. But San Francisco was fifth, Boston was seventh. I was surprised some of them were way down on the list. Like Tokyo was way down the list.

Anna Kirk: I don't know if I trust it then.

Lauren Slyman: Yeah, I see Tokyo being… I mean, just living there, it's such a dense place to live. I'd be curious what the different weights were.

Rob May: I don't think I trust San Francisco at five. That immediately invalidates the list. Cool. And what do you think London has that New York doesn't have?

Lauren Slyman: Clean alleyways.

Anna Kirk: Yeah. I really appreciate London because it's such an international city. I feel like New York is very American, and you go to London and it feels so much more international and so much more welcoming in that way.

Lauren Slyman: You know what's actually better in London, and Europe in general? People talk about the food in New York, and the food is really good here, but the average food places in London I feel like are better. If you're eating processed food, they ban a lot of the stuff that you can put in food here, and I think that's part of the reason it probably tastes better there.

Anna Kirk: Yeah. And all of the non-British food in London is fantastic. My favorite thing to do when I get to London is immediately go get some chicken tikka masala.

Lauren Slyman: It's so good.

Rob May: That's awesome. All right. So if you're listening and you're in San Francisco, move to New York. That's the lesson here. All right, speaking of moving to New York and building companies, our guest today runs a company called Quotient, which he'll tell you a little bit about — an AI-powered marketing software stack. Marc Ferrentino, welcome to the podcast.

Marc Ferrentino: Oh, thank you very much. Great to be here.

Rob May: So I want to talk a little bit about… we've covered a lot on this podcast the wrong ways that people use AI. And you guys actually go into this on your website. You talk about how a lot of these marketing tools are just ChatGPT bolted onto a CMS. I actually just wrote a whole post on my Investing in AI blog about the idea that people are AI-washing things, right? They're not really rethinking how to use AI in this new era. So tell us a little about what's wrong with just bolting ChatGPT onto a CMS. How should people be thinking about it, and how do you think about it?

Marc Ferrentino: Yeah, I think there's certain things that the harnesses do really well — the GPT harnesses as they are now. There's Rockbot and there's Muse, and there's so many new harnesses coming out on a regular basis. Those are great places for ideation. They're great places for planning. They're great places for doing certain things. But a lot of marketers try to move their entire marketing workflow into one of these tools, and they realize that at the end of the day, they're left with a bunch of markdown files. And you need to actually have a system that really does the work, ultimately, right?

So when you pair one of these harnesses with an AI-first back end, you actually get the entire life cycle together. Slapping an MCP on WordPress — it does something at some level, but it's even better to have something that's a little more modern, that has taken into consideration the amount of iteration you can perform and the amount of experimentation you can put out in marketing now. Marketing is really experimentation, for the most part. You're just trying a lot of different things and hopefully one of them catches. There's no lead magnet or marketing program that runs forever. They only run for a certain period of time. It could be a day, it could be a year, five years, but eventually they come to a conclusion. You need to always be experimenting and always be trying. And that's really what is kind of special right now: you can do it quickly.

Rob May: It's pretty cool.

Anna Kirk: How do you think about that across… So Marc, I'm not an expert in this space, but I do know that one of the challenges marketing teams can struggle with is just the pure level of fragmentation across their tool stack. How do you think about solving that problem at Quotient, and what does that value capture look like?

Marc Ferrentino: Yeah, it's a very specific approach and a very opinionated approach. We'll see if we're right at some point. There are so many tools and they all have different data dictionaries. They all have different ways of expressing things. I mean, what is a campaign across each one of these tools? It's a completely different thing. They have different definitions. The analytics in between these things are completely different. And one way to try to solve that is to create a layer above all of those different tools that collects all the different data and tries to normalize it. And that has been around for a long time.

Anna Kirk: And Marc, I'm absolutely no expert in this space, but one of the complaints that I do hear regularly from marketing teams is how much pure tool sprawl there is across the martech stack. So I'd love to hear more about how you aim to solve that problem at Quotient, and what you see as a core blocker that's holding teams back there.

Marc Ferrentino: Yeah. I've watched marketing teams for decades rebuild the marketing stack over and over and over again, and piece it all together and try to wire it all together. And the biggest challenge they have is that every tool has different APIs, different data dictionaries, different ways of tracking things. It's what makes it so hard to get any sort of visibility from marketing.

The way we approached it this time at Quotient is that we wanted to build a unified platform — one that could send email, post social posts, post blogs, do AI search, all in one platform on a common data dictionary. So consolidation is how we solve this. And I think why it's much easier to solve this today is because the price of software is just much cheaper. You can develop big, broad platforms nowadays in a way that you couldn't have years ago. It just would have been cost-prohibitive for any organization. And so what we've delivered now is a common platform with a common language, a common analytics layer, that just makes it really easy for teams to get rolling and do the thing they want to do: market.

Lauren Slyman: So what I've been curious about lately is how companies are measuring productivity as tools get more autonomous — starting with some sort of baseline and then measuring from there. How have you been able to do that within Quotient, in terms of measuring the progress made from the baseline the customer starts at to once you're involved?

Marc Ferrentino: Yeah. So typically we're working with one-to-five-person marketing teams who are pretty much underwater 24/7. They're just trying to get the basics out the door. And so anecdotally, most of our customers are now finally able to do broad-based marketing programs — weekly campaigns, a social media drumbeat, newsletters — which would normally be an impossible task for a small team like that. So just the sheer output of that team is much, much better, and that's one way we look at it.

The second way we look at it is this stat called time spent marketing. There's this great stat in the sales community called time spent selling, and that's represented by the amount of time that a rep is in front of a customer. We've been working on this idea of time spent marketing: the amount of time that a marketer spends making decisions, and not sort of in a login in a tool, right? And what we're seeing, when we've done a handful of time-and-motion studies for marketing teams, is they're spending like 60% of their time on just ops tasks. So even if I can get them a fraction of that back — because they're not worrying about that, they're not tracking UTM tags, they're not QA-ing everything, they're not having to schedule anymore, they're not having to move things between 10 different systems and have a million spreadsheets open — all of that is gone. They can actually spend time focusing on the thing that matters: the positioning, the messaging, the content calendar, the strategy. So that's typically how we look at things. If we can give that customer half their week back, or in essence double their productivity, that's a pretty big win.

Rob May: I think you're the first AI company I've heard of doing a time-and-motion study. That's like an old-school manufacturing thing, man. So that's kind of cool.

Marc Ferrentino: It sounds fancier than it is. You kind of just sit behind somebody and write down what they do all day.

Lauren Slyman: I imagine also, thinking about a marketer's workflow, so much of this is iterative too — being able to capture some of the more fragmented review points that might be nonlinear, and thinking through how you actually capture time spent when you're waiting for legal or communications teams to sign off before shipping something forward. So I'm curious how you saw that reflected in this study in particular.

Marc Ferrentino: Yeah, there's certain things we can impact and certain things we can't. Things like review we weren't able to really do much on, because that requires a human being somewhere else to act on something and respond to a task or a need. Now, maybe if they were also using their own companion AI tools, then the two together could potentially accelerate that part of the equation. But that's not a huge part of the marketer's life. I think it's their most frustrating part, but it's not actually a huge part of their day — waiting for legal, or waiting for that level of sign-off.

From the standpoint of getting internal review — for example, if you're writing a blog post on behalf of your CEO and you want them to take a look at it — we added sharing tools that put that in front of the individual reviewer. We like to say the AI can become a digital nudge, something that just bothers the person to review it. But beyond that, as of now there's not much we can do. So even that, though — the other side, I think maybe—

Lauren Slyman: Sorry, go ahead.

Marc Ferrentino: Well, even that is very helpful. The other thing I wanted to point out was that when you're writing on behalf of people and getting the individual author's sign-off — and it's very common for a marketing team to write on behalf of the CEO or an executive, whether AI is helping or not; that's a practice that's been going on forever — what I think AI has helped with is that the versions being written on behalf of that person are much closer to that person's writing style. Tools like Quotient, or the simplest version of that, a writing skill in Claude or GPT, really help at least create a version that looks similar to the way the author might actually write something. So ultimately fewer revisions are coming back, fewer round trips are happening inside the normal process of approval.

Rob May: So what's going on in the marketing AI community from the perspective of adoption? Because if you roll back to 2022, 2023, it seemed like that was the main use case, right? Write blog posts, make images for marketing content, whatever. And that seems to have, at least in the public consciousness, died down, and it was coding, coding, coding, building apps. And now that seems to be dying down as you're hearing about these more agentic use cases, non-coding agentic use cases. So what is the situation? Are most people in marketing using AI? Has it stalled out? Is it still getting better and being applied? How are people thinking about it?

Marc Ferrentino: Yeah, I think everyone is using a little bit of AI. Everyone has definitely had a Claude or GPT session to help them write something, write copy. That's your basic — I'd say that's probably 90% of the industry; that's where they are. A few people have built out more sophisticated pipelines, where maybe they're doing some research and then it's presenting them with a writing sample, usually in the form of a markdown file. That's getting a little further along, but they still have to cut and paste it into the system that's going to execute and deliver the marketing. They still have to do all the manual work around setting everything up, scheduling it, the tagging, the whole nine yards. So it's really not saving them a ton of time.

And there's some people who are just naturally gifted writers who would rather write from a clean sheet of paper than have an AI give them a version of it and then edit the version. Those people are definitely not using very much AI. They're just good writers. They don't need it. It actually slows them down. One of our customers said to us, "All it does is just move the work." So instead of writing an original copy of something, I'm just editing something — still the same amount of work for that person, right? So we haven't gotten into the stage yet where there's full end-to-end or full campaign-cycle automation. That's a big piece of it.

The second piece, and I think the biggest piece we've learned over the last few years, is that AI is quite terrible at marketing. It is really good at marketing ops.

Rob May: Yeah.

Marc Ferrentino: I mean, how many times have you seen it write something, and you read it and you're like, who would talk like this? Who would say this? Or it just feels so awkward in how it's presenting whatever you wanted to write. And so in many cases what we recommend is to share your thoughts, share your ideas, use it as a thought partner, a writing partner, but don't have it just write for you blindly. That is an absolute recipe for disaster for your marketing. So it really becomes a tool for empowerment for the marketer, less a replacement for the marketer — which is what I think most people thought it was going to be a few years ago. I think I can say with 100% certainty, as of this moment in time, it is not replacing marketers. It is just making them more effective.

Rob May: I started to try to use Claude to write more of my own posts, but it had these same lyrical patterns that drove me crazy. It always likes to put the last paragraph — it always likes to call it "the bottom line," and then it says whatever, right? And it always likes to say things like "the honest truth," "the honest whatever." I was like, "Hey, you're overusing the word honest here." But it's not even the word honest. There's something about the cadence that is so dramatic. And I think you're completely spot on, Marc, that people can smell that. People don't write in such a pithy way, and it'll overuse present tense a lot.

Anna Kirk: Yeah. I feel like you can definitely tell. But then on the other end, in my own writing — and I'm sure in many cases, especially for campaigns — you're reviewing over and over again to make sure that there's no hint of what could sound like AI, and changing your own writing to match that.

Rob May: Yeah. You know what we need to do — we won't be able to do it on this podcast — but let's have a show where we have humans and AI compete on some of these tasks. Let's get all our friends and the people we know in the community to suggest the best marketing wordsmith person they know, stuff like that. I want to get a comedian on and have them make jokes on topics versus an AI. Maybe this will be an annual thing, and we'll measure how good the AIs are versus the people.

Anna Kirk: I like it. I have a good comedian for you.

Rob May: Do you? Okay, that's good. Marc, here's what I've always wanted out of AI marketing. Do you think we're going to get to the point where I can come to a tool like yours and say, here's what I do — here's my website and everything you know about it — and I need seven customers this month, and you quote me a price to get those seven customers?

Marc Ferrentino: You've been able to do that for a long time. You don't need AI for that. How many outbound firms have reached out to you?

Rob May: But I'm saying with high probability you could do it. You would take the risk that at that price you can get me seven customers, right?

Marc Ferrentino: Yeah, you're talking about outcome-based marketing, or outcome-based software in general — a way of pricing that could apply to marketing or really many other industries. It's hard to say that that's going to be a thing, because as soon as one area gets over-optimized, it ceases to really exist.

Rob May: Yeah.

Marc Ferrentino: Meaning, like what's happening with outbound email right now: it's basically useless at this point, because it's been so automated and so overdone. And anyone who's promising you anything in that arena — it's maybe wishful thinking on their part. So I don't know, because the thing that we all respond to is opinion, is taste — a point of view. And I guess maybe one day the AI can get sophisticated enough that it could figure all that out. I just don't know when that's going to be. I always take whatever Elon Musk says — "we'll have superintelligence by next year" — to mean we're going to have it 10 years from now. So maybe 10 years from now, we'll have it.

Rob May: You and I share a lot of common views on this, I guess.

Marc Ferrentino: Well, I mean, he told us we were going to have self-driving cars in 2007.

Rob May: Yeah, man. And some of these lunatic AGI email groups I'm in — these guys are like, it'll be so smart it'll be able to convince any person in the entire world of anything. And I'm like, come on, man, there's no way.

Marc Ferrentino: No, because it was created by humans.

Rob May: I know. And that whole statement is just not possible — to be so smart that you could convince any human. Some people are just never going to believe certain things, no matter how persuasive you are. It just shows me these people don't understand how humans work.

Marc Ferrentino: Well, also, what I guess makes sense is that the most effective forms of marketing right now are field marketing — experiences. And I think influencers are also very big. But now we're starting to see the automation of influencers: basically AI characters, AI casts and actors that are creating their own channels and putting things out. So how long before that gets saturated and stops working? And that's back to the crux of marketing: things work for a period of time until they don't.

Lauren Slyman: Yeah. I think the theme that you're hitting on, and that I'm seeing on the more optimistic side of AI in most industries, is what you were saying about freeing up operational work so you can have more time for these field engagements and this one-on-one personal contact.

Marc Ferrentino: Yeah — as long as we're making the buying decisions. If AI starts making the buying decisions, everything I just said has to be reset. But as long as humans are making the buying decisions…

Anna Kirk: That's actually a good segue into one of our articles for the week, around TikTok rolling out an AI shopping assistant. How do you all feel about that? Marc, did you hear about this?

Marc Ferrentino: I did. Yes.

Rob May: Yeah. So we talked about this a little bit on the last show that I did with Stephen Messer. One of his things was — so let me back up. Neurometric is my fifth company, and typically when you've done a lot of marketing, you start to get a feel for numbers. When somebody says numbers, you go, "That doesn't match up," right? You see certain people coming to your website; you have a band of conversion rates that you expect at different levels. And we saw this pattern where lots of people are coming to the website, they're clicking around a lot. They're not looking at everything like a scraper would do, but they're clicking down certain paths, and they're not doing anything. They're not filling out a form. If somebody clicks on a bunch of your content that's all related, you expect they're going to fill out a form or email you, or at least some percentage of them do, and none of them are. And so we think it's agents. We think what's happening is there's agents that are coming out and poking around and doing some stuff, but then they're not authorized to do anything, so they're leaving.

So we started talking about this idea: when agents start doing a lot of this on our behalf, how do you market to the agents? They're much harder to market to than people, because with humans, everybody has something that they want — how they want to feel, how they want people to think about them. There's something more than just logic that you can tap into, and they have intents and all these kinds of things. And these agents are just information gatherers. It's a marketer's worst nightmare. What are you seeing on this, Marc? What do you guys think?

Marc Ferrentino: I am absolutely seeing it, yes. There's a reason why AEO is so hot right now. You've got companies like Profound, like AirOps; you've got HubSpot, the legendary giant of B2B marketing — if you go to their homepage, it's all about AI search. Same thing coming out of Adobe, coming out of Salesforce. So it's a hot topic for this exact reason: how do we get the agents to know about us? Step one is simply, do they have the information about us, so that they can make the informed decision or help make the informed decision?

And luckily, it used to be all about getting your information into the weights of the model. That is not really the game anymore, because when we went from LLMs to agents, agents actually have tools, and those tools can search the web, right? So the joke of it is that SEO is kind of back to SEO, in that when you ask it a question, the first thing it's doing is searching, just like a human being would be doing the searching. So I would say SEO and AEO are just as important as they've ever been, but it feels right now like, if the agents are making decisions, it might be a little more important than it's been historically.

I think for certain businesses, inbound marketing via search — intent-based marketing — has always been super powerful as a channel. And it's even more so now, because think about it: you're asking an agent to go research something. That's your intent. So your intent goes from the Google search box to the agent now, and we want to capture that intent wherever it may be. Could be on TikTok, could be inside of an agent, could be in the Google search box. We're just looking for intent, and we're going to go wherever it is.

Lauren Slyman: I would love to see the telemetry data on that.

Marc Ferrentino: I can tell you from what we're seeing on our own website: about 35 to 40% of our traffic is coming from Claude and GPT.

Anna Kirk: So when are you going to invent some robots for field marketing?

Marc Ferrentino: That's the one thing I can't do.

Anna Kirk: I feel like that's the next thing.

Marc Ferrentino: We actually have an events marketing component to the platform. But what I say to people is, "Listen, I can handle all the digital parts of setting up that dinner, that webinar, that trade show — the before and after outreach, making sure that your dinner's full, your webinar is full, or your trade show is booked from morning till night. But I can't help you in the booth. Not yet."

Anna Kirk: We'll get there with robots, maybe.

Rob May: All right. Speaking of models, and the fact that New York is the number one city in the world: New York now has a frontier model company. I mean, we've had one for a while, but they actually released their model — Reflection AI. It's like a 500-billion-parameter model. Do you guys know the story here? These guys were smart. They came out of DeepMind or OpenAI or something like that, and they were going to build some other tool, and they met with Jensen Huang, and he was like, "The US needs an open-source US model. You guys should build it." Now, since then there have been some other companies. Arcee is one. Meta's back in the game now with their models. But this one sounds like it's pretty interesting. It's been optimized for coding and other stuff like that. These guys have said they'll come on the podcast at some point — I'm in touch with their PR team — so hopefully we'll get to talk to them about it. But did you guys hear anything about it? Any early reviews, or has anybody tried it out?

Anna Kirk: I've not tried it out. I have heard about it. I do think it's very impressive, and it touches on an important point about what you could call a battle of sorts, or competition, over who wins this race. A lot of really strong open models have come out of China, and they're cheap. So I think this is a very interesting competitive advantage we have here, and I do wonder if open-weight models will become the new determinant for whoever wins — quote unquote — whatever this hypothetical race is.

Rob May: Since you're our resident expert on AI details, because you're writing a book on how this works, do you want to explain to people — this is a mixture-of-experts model instead of a dense model. Do you want to explain what that is?

Anna Kirk: So I have not gotten to that section yet. I will come back with a better explanation. But yeah, I do think it's hard, at least for me, not to draw parallels between this and — not to get too dramatic — the space race. It does feel a little bit like an arms race, where the ultimate outcome is probably just going to be pricing pressure and commoditization, and that is a good thing for consumers. Marc, I'm sure you see this as well: people are starting to become very aware of what their token usage is, and if they have a viable open-weight option, they're going to take it. And I think increasingly harnesses are building in the capability to do cost-aware routing.

Marc Ferrentino: Yeah, that's something we… we're a multi-model harness ourselves, and people were blowing themselves up with Fable, and they were blowing themselves up with the old Opus pricing. They became very aware of it because the prices were — they were token-maxing, they were going crazy, and definitely having an issue around spending too much there. The funny thing is the prices have come down so much. We're just in the process of implementing Kimi K2 as one of the models in our harness, along with GLM 5.3, and we waited so long for Kimi that now Kimi is not a cheap model anymore, because Sonnet and Opus have come down so much. It's actually somewhere between the price of Sonnet and Opus. It's not even the cheapest model anymore, and it's a Chinese model, right?

Rob May: Yeah, we track — because we move people to small models and route between those — we track the prices of all the models and how they jump around. What you're ultimately going to have to do is make it easy to switch between models, because this is going to be an ongoing thing. And the pace of model releases — I mean, just this week, it's Tuesday evening, we're shooting this, and in the last couple of days you've had Gemini 4, you've had a new model from Mistral, you've had Reflection. The pace is crazy. And I think that's going to continue, because just like with human beings, and how humans are specialists attuned to certain things, that's where we're going to go with models. Even though you could build one model that was the best at everything, it would be slower to run, bigger and more expensive. So — and I think maybe you mentioned this earlier, Marc — you're going to have a couple of really big models that do everything, then core workhorse models that are specialized to common big use cases, and then a bunch of small niche models for all the little repetitive tasks that you just want to get done as cheaply as possible.

Marc Ferrentino: It's a normal distribution.

Lauren Slyman: Yeah, I was about to say: mimicking the human hierarchy we have currently. Statistics nerds.

Rob May: Well, all this talk about token-maxing leads into one of our last articles, which is The Information just talking about tech companies — particularly big tech companies like Meta and Microsoft — weaning themselves off of Claude because it costs so much. Are they? You're in there. Are they really weaning themselves off?

Lauren Slyman: I knew I was going to get asked about this.

Rob May: You can say no comment.

Lauren Slyman: Honestly, I don't even know. I know as much as everyone else in the public and at this table. Who knows? But I assume, just because of the OpenAI relationship, that it's probably more economically efficient.

Rob May: Yeah, I think it's very telling, and I do think it might be part of the reason why Anthropic and OpenAI are trying to figure out how to tell this part of the story in their IPOs, right? Because their whole thing has been that they were the only games in town, and now it looks like everybody's going to have a game. So how do they differentiate and keep customers? I mean, it is weird — Marc, how long have you guys been working on Quotient?

Marc Ferrentino: A little under three years.

Rob May: Okay. So when you guys started, the point of view in the world in 2022 and early 2023 was, "Game over, OpenAI won. They're the AI juggernaut." And then Anthropic spun out of there, and people were like, "Come on, man," and then, "Okay, so there's going to be two, right?" And then you had the DeepSeek moment in early 2025, and people were like, "Oh, holy shit. There's going to be legit open-source competitors soon." And it just keeps going. I think it's surprised everybody, because the pitch of these companies in the early days was: it takes hundreds of millions of dollars to train a model, there'll only be a few companies that can do it, some won't get it right, there's going to be one or two winners. And now it looks like there's going to be 15 model companies.

Lauren Slyman: This reminds me of an article that we read and reviewed on one of the first podcasts — that you sent. I can't remember; maybe you remember the name of the author. "AI Will Not Make You Rich." Do you remember that?

Rob May: Oh, that was — who's the guy that teaches here? He's a professor here. We'll edit that in, or put it in the show notes.

Lauren Slyman: Jerry Neumann. We should have Jerry Neumann on.

Rob May: Oh, I would love to have Jerry Neumann on. That would be so fascinating.

Lauren Slyman: Yeah, he was right. We're seeing more and more models coming out that are incrementally capturing value, but less and less and less. Ownership isn't accruing to one spot. And I think there's an interesting vacillation back and forth when you talk about new tech adoption — it makes people feel comfortable to go to extremes. Either you're going to catastrophize when DeepSeek comes out, or it's like Y2K and you're worried that planes are going to crash in the middle of the sky because some computer thing hijacked it. I think it's really unsettling to people when the answer is somewhere in the unsexy middle: people are going to continue adopting things and using things the way that they always have, and there will be segments that are profitable and segments that aren't.

Marc Ferrentino: Precisely. And all of us who are entrepreneurs are betting on which segments we think those are. Will the value accrue to the harness? Will the value accrue to the model? Will it accrue to the hardware, or someplace else that we don't even know yet? I'm not sure we know, and that's fine. Well, we would be in the harness arena, so that would be great if that's where it was — good old-fashioned software. But I do remember, back in March of this year, it felt like it was game over: Anthropic was the winner. Every day I'd wake up and there would be some announcement about Anthropic for law, or Anthropic for real estate, or Anthropic for something, and everyone was like, "Game over for that industry. It's over." And I'm very happy that that reality did not stay the reality, because that was a terrible time, in my opinion — in AI, in software, in just our human existence — to have everything go to what looked like this one company. And it's not that I'm rooting for Anthropic to fail; I'm rooting for them to have proper competition.

Anna Kirk: Pick a lane.

Marc Ferrentino: Yeah. When Elon went back in and started working on Grok, and Grok 4.3 came out, I was like, "Okay, great. That's a second one, or a third one." And then the open-source models came out — GLM and Kimi came out — and I was like, "Okay, we've got ourselves a proper battle here." It's a proper industry, with different players all fighting for supremacy, all driving the costs down and driving the value and the quality up.

Rob May: Well, and that's what makes me not worry so much about whether this is a race with China to get to recursive artificial superintelligence. All these people are within days or weeks of each other in terms of how good they are. Let's just do the thought experiment. Let's say you're the first AI to start to be recursive. You still have to make some probabilistic guesses about which paths to investigate to improve yourself. It's like playing poker, right? Sometimes the low-probability things win, and that could happen in this field as well. So I think there's plenty of time for everybody to catch up. I don't think it's a winner-take-all game at all, even between countries.

Lauren Slyman: Yeah, and it shouldn't be. It reminds me of a lot of the ethos that we saw around Bitcoin versus Ethereum in 2018, where everyone was like, "Is ETH going to win out, or is Bitcoin going to win out?" And flash forward almost 10 years, and they are…

Rob May: It has no use cases. They all have no use.

Lauren Slyman: I won't say that. That was a great time on Reddit, though.

Rob May: Yes. Good — last article that we have for today, since we're talking about New York, and since we're talking about China and US AI. We've talked several times on here about the idea that the US's immigration policy is hurting us in this area, because we should want everybody who's smart and working on this to come here. There aren't enough people in the world. And it turns out China has been trying that and has not been very successful. The New York Times reported that about 30 times more people come from China to work on AI here than do the reverse. Does that make you guys feel better about the race?

Anna Kirk: I'm surprised. Was this a problem for them? I feel like they're pretty advanced, with their models, and they have amazing talent already domestically. So I was really surprised reading that article — if it's the pure number of people. I am curious how many times more people go to China for work for anything. Do we know how this compares to other industries right now? It could just be that's the average, right?

Marc Ferrentino: I didn't know that anyone went to China to do AI research or work there. What I remember is that a few years ago, at a company I was with, we had opened up an office there, and then their policy changed, and almost every software company — most businesses — wound up packing up shop and leaving. It seemed like they didn't want us. They weren't interested in any of us coming over. So this idea that Xi Jinping has invited 100,000 students to come over — I was like, where'd that come from? That's a big, big change. A big 180. But okay, sure.

Rob May: Yeah. It's going to be interesting to see. I have this mixed point of view on how much this is just a numbers game. Because I sort of simultaneously believe a little bit in the great-person theory of history — that some people come along and they just have the right combination of talents and experience at the right time, and they move everything forward in a way that other people wouldn't have. But I also believe that, probably with the exception of special relativity, everything else would have been discovered about the time it was discovered anyway. That was the one thing that just seemed like a random leap out of nowhere.

So I think maybe part of what they're trying to do is get those… You guys have worked at companies, right, where you have some catalyst person, and this one person on a 40-person team makes the whole team 35, 40% better. It's something that they do — they just see things, they can pull things together. It's not always that they're the smartest or whatever. I assume working on AI research is the same way. You probably have those special people who you can drop in different locations, and they just vibe, and they go, "Yeah, let's go with that idea." So maybe that's what they're doing. They're trying to get those catalysts over.

Marc Ferrentino: It's always a numbers game, right? The more smart people you have working on a problem, the greater the chance you'll find that next great person who will move it forward. I can tell you right now, if you have a fraction of the number of people working on something, it's going to be much harder to find that person. So it's never bad to expand the pool of people focusing on big problems. And the only reason that students go back and forth is to share information, to share knowledge. So great, let's do it. And honestly, the Chinese models are pretty incredible. If you've used Seedance 2.5 for video, it's insane. And I have no issue whatsoever using a Chinese model.

Lauren Slyman: Yeah. So we're all going to use whatever works.

Marc Ferrentino: Agreed. Whatever the best model is that helps us get the job done, we're going to use. I don't see a world where you're being patriotic about getting your job done and picking a model because it comes from a certain place or not. So hopefully, the more we share back and forth, everyone has the same thing, and then there's no advantage anywhere. The advantage goes to the individual who's using the model.

Rob May: Yeah. So Marc, for your context, at the end of every show we have some kind of favorite New York X — favorite New York burger, favorite New York pizza. I thought of this mid-episode today, but have we done favorite New York memory? One of my favorite movies is Serendipity. Have you seen it? John Cusack just meets Kate Beckinsale, and they decide to go ice skating, and she's like, "Favorite New York memory?" and he's like, "This one's climbing the charts."

Lauren Slyman: See, there we go. So it's this one, Rob. It's the one we're making right now. Shooting shows with you guys.

Rob May: All right. Well, I guess — do you have a favorite New York memory, Marc?

Marc Ferrentino: I mean, I lived in New York for 23 years. I'm from the area, so I've got a lot. Everything from the Knicks' Game 7 in the early '90s to walking my kids to school, to Grace Church School, down on Fourth and 11th Street. Every single company I've ever started has been in New York. So I've got a deep history there. I'm trying to think of one of my favorite memories. God, it might have been when the New York Football Giants won the Super Bowl in 2011, and just being there in the crowd. I wasn't there for the Knicks winning this year, so I wasn't able to be in New York to feel the New York energy when a hometown team wins. The last time I was in New York for that was the last time we won anything — which was a long time ago now. And just being there with the fans, everyone coming together, everyone just being New Yorkers, not being in our little tribes — it's always a great moment. And I talked to a lot of people about the Knicks when that happened recently, and it was a great moment for the city. Whenever we win, the city comes together in a special way.

Anna Kirk: Yeah, I was about to say that. I feel like if there's one special thing New York has that many other cities don't, it's that we're great at collective celebrations. I think one of my favorites was when Biden won the election against Trump in 2020. I don't know if any of you were in the city at that point, but it was still immediately post-COVID, and a lot of people weren't vaccinated yet. So the ability for the people who had endured lockdown in this city, and seen a lot of elements of their city taken away, to then be able to just walk the streets and come out and say hello to everyone — the weather was beautiful, and it just felt like this big parade and sense of exhaustive celebration with everyone in the city. That just felt like the coolest moment to see in the middle of what was otherwise a pretty [difficult] time.

Rob May: Thought you were going to say Mamdani's election.

Anna Kirk: Oh my god. Well, that's a close second.

Rob May: Mine was definitely when the Knicks won this year. That was my favorite night in New York so far. It was awesome to see. And despite what the national media profiled — a couple of bad apples — everybody I saw was nice and helpful and having a good time. It was awesome.

Anna Kirk: Yeah, the media definitely skews it. I get a lot of texts about it from people who don't live here, and I'm like, we're doing great. We're having a great time.

Rob May: Yep. What was your moment, Lauren?

Lauren Slyman: Oh gosh. I'm just going to go with the Knicks as well. I wasn't here for when they won — I was on a plane to Maui. But before that, every game, I felt like it was almost even more exciting, because of the excitement leading up to the final game. It reminded me a lot of college. It's this single moving organism, this city and the people in it. During the playoffs, people walked by before and after the games yelling to each other, "Hey, go Knicks," high-fiving and stuff. It was a lot of fun.

Anna Kirk: Do you remember there was someone in the West Village who just decided to take a projector and put it against this blank wall that was right outside their apartment? I've never seen it so packed.

Lauren Slyman: Yeah. It was a pretty special moment.

Rob May: The random things that people do in New York are the most awesome. That's what I love. Well, Marc, thanks for coming on the podcast today. If you're interested in the stuff that Marc talked about, you can check out his company at getquotient.ai. And if this is your first AI in NYC Show, go watch some other shows, rate us, leave us a stupid comment on TikTok — that seems to be a thing now. And we'll see you guys next week.