Agentic Network Effects and Defensibility —Fireside Chat with James Currier

Julia Nimchinski:
Now we have a very special guest. Welcome to the show, James Currier. Excited to feature you. How have you been, and what’s in your AGENTIC OS?

James Currier NFX:
What’s in my Agentic OS? Oh, all the things. We’re, I wasn’t ready for that question. We can talk about that later. Why don’t you ask me some questions?

Julia Nimchinski:
Awesome. How about just a quick introduction? I’m happy to do it, but .

James Currier NFX:
Oh, an introduction of me! My name’s James Currier, and I live in Palo Alto, and I originally grew up in Boston. I went to Exeter, Princeton, Harvard Business School, did all that, did venture capital in the 90s, and then started 4 venture-backed companies where I was the CEO. They all did well.
So I got the right to start this venture capital firm in 2015 called NFX, which stands for Network Effects, and we’re now, $1.6 billion dollar seed fund that, you know, we’ve got 5 partners, equal partners. We’ve got offices in Israel, Palo Alto, San Francisco, and we make investments of, you know, sort of 3 to 4 million.
We have about 16% average ownership, and we’ve done about 220 investments, and Now, all is going well.

Julia Nimchinski:
Moderate to future you, I have millions of questions, and can get… can’t wait to get into it.

James Currier NFX:
Great.

Julia Nimchinski:
I’d like to start off with the Bible, or the NFX one. And your NFX methodology, network effects methodology. Yeah. Just… to add the nuance of AI, I’m just curious, where do you see the place of AI? Is it a node, is it a link, or is it just an entirely new network architecture?

James Currier NFX:

Yeah, so the history of, sort of, understanding how network effects work, you know, in the 90s, there was Bob Metcalf, and in the 2000s, there was… Tom Eisenman, and then I’ve sort of inherited the mantle of the person who studies this stuff the most, so if you go to NFX.com, there’s tons of content about how network effects work, and there’s 17 different network effects, and… You know, these have been responsible for the multi… you know, Decacorns and Centacorns and all that stuff, and the trillion dollar companies that we see around us today.
And so, they’ve been very powerful. Microsoft and every company you kind of really admire has these network effects. And so your question’s a good one, which is, given AI, does it really change how these network effects work? And the answer is no. It doesn’t. The way the math of defensibility around network effects works… now, let’s be clear. Network effects does not mean viral effects.
Viral effects are the way you use your current users to get you more free users. That’s a viral effects playbook. That is not network effects. Network effects are about retention. Network effects helps you retain your users and to capture value that no one else can then provide, alright? So the way those mechanisms, those 17 network effects work, those are still working the same way.
What AI does change is that you can now add in cognition. And action, somebody reading, somebody writing, somebody paying, somebody… it doesn’t just have to be humans anymore. It can now be nodes that are made up of AIs.
And the difference is that there’s only, you know, one James Courier, but now we could create a thousand AI James Couriers, or 10,000, or a million, and we could do that for every human. And then we could create other non-human humans.
So the proliferation of nodes on your networks can be much bigger, much faster, and more replicable, and so the network effects Are gonna be different in the sense that you’ll have to adjust to that element to it. But the fundamental 17 network effects remain the same.

Julia Nimchinski:
Love it. James, you spoke to the evolution of the connection between nodes. So we started with one-to-many, broadcasting, one-to-one. And now we’re seeing some newer architecture with MoldBook and others, where it’s centrally, correct me if I’m wrong, but it seems like one-to-one-to-many were, like, siding to side, agents collaborate, upvote, and then human-observe.
What are your thoughts on this one? And… and yeah, how would you call it?

James Currier NFX:
Yeah, so MoltBook, is still, a direct network effect. The math of it still persists, it’s just that the order in which the nodes are taking action has been reversed, from humans doing stuff, and then we put AI agents to watch what we’re doing. we reverse it the other way. The question MoltBook has to figure out is, is there any defensibility there?
Because I could create a MoltBook and make another 100 really interesting agents, and then other people could add in thousands more agents. He’s gotta figure out how to make it a defensible network effect by adding in sticky nodes, which is humans.
And and I don’t know if that’s happened yet, I haven’t been tracking it on a daily basis, I know it’s evolving quickly, but, I don’t know that it’s a different type of mathematical architecture, it’s just the order has been switched, which I think is interesting, and certainly we’ve been talking quite a lot about it.

Julia Nimchinski:
Speaking of what happened, so with Tickle, because you’re not only, you see, you are also a 5X founder, super successful one, so one-fourth of the internet. were captured by Tickle at the time. If you had access to AI back then, would that change anything? The speed, the penetration, or… what are your thoughts here?

James Currier NFX:
Yeah, so Tikko was a company we started in 1999 to do self-assessment testing and then social networking around people’s interest in themselves. It’s our number one interest, right? You know, some people might be interested in Ford, some people might be interested in, you know, House… you know, Game of Thrones or whatever, but everyone’s interested in themselves.
And as a result, we registered 150 million people out of the 600 million people on the internet at that time, and that was because everyone sent around these test results to say, this is what I am, what are you, let’s share and compare.
And it, you know, eventually, you know, other people sort of copied it and whatnot, but… You know, we sold that in 2004 to Monster.com, but it would have been different if we had AI, because instead of spending 5 years building 450 tests. We could have built 450 tests in a week. And then we could have, you know, we could have written them in more depth. We could have created more graphics.
We were doing it all by hand, and so it rolled out, you know, over time. Now, not only would we be able to build those 450 or we could have done 1,000 tests, but we also could have used AI to analyze the data, and done a lot more with the data. Then we were able to. We had this huge trove of 25 billion questions answered by people.
We knew if they liked their mom or their dad better, we knew if they liked, you know, pickup trucks or minivans. We knew their favorite colors, we knew all sorts of stuff, but we weren’t able to do much with it, because it just would have taken too many human hours to call through the data. So, yeah, it would have been very interesting.
It would have been supercharged, is basically, what it would have been had we had AI.

Julia Nimchinski:
Speaking of people and language, I know you’re really big on this one. So, I’m paraphrasing here, but you mentioned, that essentially as a founder, you have to figure out the language, and then build exactly that. The other way around. So I’m curious now, with this trend of agent buyers, in B2B and B2C early innings. Would you say that AI has its own language?
And as founders, should we figure it out?

James Currier NFX:
Yeah. So, yes, AI will have its own language, it will develop its own language, that’ll be much more shorthand than English. And we will need to build translators so we can understand what’s going on. you know, I’m not sure it’s going to matter too much to the AIs that we understand or not, but it’ll certainly help us debug these systems as we continue to sort of overview what’s happening.
No, I think we should be developing, or I think we should allow it to develop its own languages. Just so that we, you know, save inference costs, and we save, transaction speeds, and… And, that’s coming for sure. But it’ll be fun for us to have translation and see what they’re up to.

Julia Nimchinski:
Super fun. I’m curious about, your thoughts on B2B in general. So we’re seeing a lot of, you know, just disruption happening in sales, marketing, CS, and essentially the functions bound. Do you think that we’ll have this division, B2B versus B2C in future?

James Currier NFX:
I think in the last 10, 15 years, it’s, you know, it’s obviously a spectrum. You go from pure consumer stuff to sort of SMB, sort of individual… individuals who are doing, you know. you know, things like Shopify, where it’s just individuals can run a store, and then it goes up from there.
So it’s always been a spectrum, and I think that we use the terms B2B to And B2C to… you know, sometimes ill effect, because many times, you know, if you’ve got heavy-duty enterprise, that’s a different type of sales cycle, and I think when people say B2B, that’s what they’re thinking of.
But there’s plenty of mid-market and SMB B2B sales, which is more consumer-y and would benefit from more consumer. And so I think our terms do obfuscate and not serve us. When, when we say this, this sort of stuff.
I do think that the… this… look, there’s an interesting article on NFX called Technology Windows, and I encourage everyone to go read it, because what it shows is over history, all technologies have these windows that open and then they close. And during the window being open, it’s a good time to start a business using that technology. But once the window closes.
It’s not so great, and there’s really no much… not much you can do, no matter how talented you are or how hard you work. And so, we did have consumer software, 1994 to 2013, and then the window closed, at least in the West. Okay, Europe and the United States. And we had maybe 10 unicorns after 2013? That’s it. And we would have 20 a year. between 1994 and 2013, and then we had 10 in over 11 years.
Now, AI hopefully will open that back up, and there’s an article called Consumer Is Back, and it talks about, hey, I could open that window again. But the same thing is happening with B2B software, which is that we had this great run. of on-prem during the 70s. We then had a flat line in the 90s, where almost nothing happened with B2B software.
2000, you get Salesforce, and you get browser-based B2B applications in the cloud, and then we had a lot of unicorns created between, you know, 2000 and 2020, but the window seems to have been closing, and the incumbents are really capturing 94, 97% of all the value in these markets at this point. And so, yeah, the B2B software window has been closing and will continue to close.
AI opens it up quite a bit. And then that will then… also, that window will then close, maybe in 3-4 years, and then software will kind of be done-ish, is my guess.

Julia Nimchinski:
Speaking of zooming a bit out of just software and B2B, I’m curious your thoughts, James, generally on network effects and the laws that regulate network effects.
I’m also paraphrasing here, but once you see it, you can unsee it, and you start to see it in politics, you start to be selective about, you know, the nodes you connect to, and the architecture that you laid out, you know, your LinkedIn or other networks. I’m curious your thoughts, so we have Sarinov’s Law, Reed’s Law, Metcalfe’s Law. What’s the next one?
Do you see it forming, or it’s still the three fundamental?

James Currier NFX:
It’s, look, I think that, you know, the other network effect, the other big network effect is the marketplace network effect, and we actually came up with our own law around it. We actually figured out the base formula, and In the NFX Masterclass on Network Effects, it’s in there. Like, we… we haven’t actually done a separate article about it, but it’s in… it’s in one of the videos in there.
It explains what… the law is around marketplace dynamics. But we haven’t seen any other laws emerge yet, so, So, but it could. It could very well. We… you know, we got a lot of these laws recognized once the internet showed up, and now that AI is here, we might start seeing some more, but we haven’t seen it yet. Maybe someone will come up with it, maybe it won’t be us, maybe it’ll be us.
We’ll see.

Julia Nimchinski:
I also call it the courier effect.

James Currier NFX:
The courier… Yeah. Yeah. Maybe. We’ll see.

Julia Nimchinski:
James, you talk a lot about Uber. And obviously, it’s hard to talk about Uber and marketplaces like Uber without their ability to asymptote. Do you think that… and do you see AI regulating that quality?

James Currier NFX:
Do I see it regulating the asymptoting? Yeah. So, so basically, I mean, in marketplaces. in things like Amazon Marketplace and whatnot, the value to user… the value to buyers goes up almost linearly with the value… with the number of suppliers. with marketplaces like Uber. the… the value asymptotes when the… when the car can come to you in 3 minutes.
Like, is it that much better to come in 2 minutes? Like, you might need to get your jacket, you might need to go to the bathroom. It’s not necessarily better to have it come faster, and so there’s… this is what we call asymptoting elements of marketplaces.
And, AI, will, I think, bring people to these asymptotes faster, just because they’ll be more efficient at leveraging a smaller number of nodes. To get people to, sort of, optimal, optimal service levels, or quality levels, or price levels. And so we’re gonna… we’re gonna get to these asymptotes faster. And of course, you know, with eBay, you do get to an asymptote at some point.
But it’s pretty far out, and I think with AI, we’ll get to it faster than eBay did during its heyday of the 2000s. If you were to start an eBay today, you would hit that asymptote faster because of AI. It’s just more efficient. Not because it’s different fundamentally, but simply because it’s more efficient with the same math.

Julia Nimchinski:
Makes sense. When people talk about AI, especially, you know, as part of this summit. There’s a lot of conversation happening, with its ability to do things autonomously. And I’m curious your thoughts on With incivility. And, just the quality of, I mean, the first defensibility, the strongest one, embedding. Do you see AI just doing that autonomously?

James Currier NFX:
Yeah. Yeah, we… we… we write a lot about defensibility, because that’s where all value comes from. Right? Retention and defensibility is where all value comes from in companies, and so we’re obsessed with that, and you can go on NFX.com and read about… just type in defensibility, you’ll get a whole bunch of articles.
What’s… the two main defensibilities available to startups are, as you say, embedding, which is like Oracle. Once you’ve embedded Oracle in your company, you’re gonna retire before you rip that thing out, and they can just charge you 25% more the next year, and you can’t do anything about it. And then the other one, of course, is network effects, which is even bigger. But those are the two.
The other two main defensibilities, which is brand and scale, those really aren’t available to startups, because they don’t have enough money, they don’t have scale, nobody knows who they are. So, we focus on those two. And, you know, B2B software has generally leveraged embedding as its defensibility. It tends to have a hard time getting network effects.
And so, what we’re seeing with AI software is that it also needs to embed itself in your operations so that it becomes very difficult to rip it out, because then you’ll lose money this quarter. And no one wants to go through that pain. of doing that to make a software switch.
And so, yeah, I think that, the land and expand of a B2B software company will be even faster with autonomous units, as the agents will say, hey, why don’t you invite this person in your organization to join us, you know, to join you on this software platform.
Why don’t you, suggest that you add this module for this group of users in your company, so that the five of you can work together on this type of a CRM versus that type of a CRM, or this type of a calendar, and so… and years ago, we posited that you’d end up with just two giant software companies. This is the further extent of the mental model, which is orange and blue.
I have Orange Software, I’m one person, I bring Orange software into my organization, I hire 3 more people, they also get Orange Software, but Orange Software watches our Slack and watches our email and watches our calendar, and then says, hey, looks to us that… looks to me that you could really benefit from a CRM. Should I build a CRM for you?
For this specific market, for this particular geography, for the way you and your co-founders work. And you’re going to end up after 10 years with this organic orange software, which will just be completely form-fitted to who and what you are doing and what markets you are. Same thing with Blue, and they’ll just compete.
So if you take that mental model, that framework, and then think about how it affects B2B, you can see that, yes, the autonomous agents of of B2B software are going to spread like… like tendrils, or like a tumor in your brain into the organization, and sort of get that embeddedness so that, you know, it can sustain itself as a software company.

Julia Nimchinski:
It’s hard to compare AI and network effects, but from the perspective of innovation. Do you see in network effects more powerful, or AI flipping network effects?

James Currier NFX:
No, I don’t see AI flipping network effects. I think that, in fact. If you look at the history of defensibility, Prior to 1994, There were tens of defensibilities. I owned the license to the port. I own the contract to the mine. I own this patent. Da-da-da-da-da. But digital removed almost all of those. you know, I’m in this geography. I am doing it in this time.
Digital removed time and space and software, the IP, the patents didn’t really work. So almost every defensibility went away, except for the four I mentioned. Right? Network effects, embeddedness, scale, and And brand. And… And now, I think AI is just reducing it down to network effects and embedding, if you’re very clever. Because the agent doesn’t care about the brand.
The agent has an infinite capacity to understand what’s talking to it. Right? A brand is useful for a human simply because we have a scarcity of ability. We have to take shorthand. Oh, Nike. That’s a good… that’s a good brand, I’ll buy that one. As opposed to unknown brand, I cannot evaluate. I don’t… I have scarcity of cognition.
I can’t evaluate whether that piece of clothing is sufficient for my needs. I’ll just trust the brand. It’s a shorthand. AI doesn’t need shorthand, so brand kind of goes away, and it’s going to have its own language. Right? The AIs will be talking to itself and exchanging information, so brand sort of falls away. Right? And then scale, AI has infinite scalability.
We just jump on the backs of AWS, or on OpenAI, or Gemini, or Anthropic, and it’s an infinite scale. So those two defensibilities go away. So we’re really just left With network effects and with embedding. as our defensibility. So.
I don’t think… I don’t… I think we’re going to be even more desperate to get to network effects and to get to embedding than we’ve been even in the past, and so I don’t think it overturns it, I think it actually emphasizes it. It’s… in economics, it’s known as a Geffen Good, but I… that’s a whole… you know, you can go research Geffen Good and see what that is.

Julia Nimchinski:
Thank you. James, on NFX, you have another article about the three waves of generative AI tech, if I’m not mistaken. And, obviously, we’re just, you know, in the very early innings. But it’s also obvious that, a PC, an iPhone. computer is now the ultimate device for AI to flourish. What are your thoughts here?
And, I mean, we know that OpenAI is building something, Enigma, but yeah, what do you think?

James Currier NFX:
Yeah, so in terms of the waves of AI, I think we are now 3 years into a 12-15 year cycle that I call More and better. So everyone’s worried about job loss, everyone’s blah blah blah. I’m not worried about it for the next 12 to 15 years. Okay? I think that we’re just… everything’s gonna be more and better.
We’re gonna have more seminars, we’re gonna have more art, we’re gonna have more music, we’re gonna have more jobs, we’re gonna have more sales, we’re gonna have more… more and better everything. After that, it’s hard to say what starts to happen.
When we get out to there, and we start looking at other interfaces, there’s clearly going to be some sort of auditory, some sort of, some sort of glasses thing, there’s gonna be some sort of implant. Right? Right now, you know, the ring, the Oura Ring can do more and more. We’re getting more and more data. You know, the watches can do more and more.
Those seem to be the form factors that could work. counterintuitively, I actually think that this interface is going to persist for quite some time. The reason is, it fits in your pocket, you can remove it from the social situation appropriately, as opposed to the glasses, where you never know if they’re recording everything.
And it carries a battery to, you know, because we’re going to do more and more AI on this thing. And so we’re gonna need, actually, bigger and bigger batteries for this. Because we’re gonna take it out of the cloud and bring it here, just do it on device. And so I think this form factor is actually going to stick with us for quite some time. In terms of the laptop.
Hard to say, but I also think it’ll stick with us for a little while, simply because It’s… it’s shaped to the human body. And at least for the next 15 years, the humans are gonna have most of the wallets. And as a result, we’ll have to cater to the humans still.
And so… but yeah, no, I think those are all potential form factors that… that will advance, but, And I’ve seen some things that are, like, patches over here. We’ll see what happens there. But yeah, there’s plenty of opportunity, but don’t expect this to go away too soon. It’s really good for a lot of reasons.

Julia Nimchinski:
James, pivoting a little bit to your, I guess, favorite topic, marketplaces, you mentioned, that As a fund, you now invest way less in marketplaces. Regrettably. Yeah. And… Assuming it’s for… because of the lack of originality and original ideas. But what marketplaces have you seen that actually fascinate you now, in the age of AI?

James Currier NFX:
Yeah. So, the way we work at NFX is we… there are some venture firms, there are a few venture firms that are very thesis-driven. We are not. We are founder-driven. And so we receive each of the founders into our office and onto our Zoom calls. sort of a blank slate, and we let them show us the way, right? We used to be founders ourselves.
I think we built 10 companies and sold them for $10 billion, so we’ve been in the shoes of the founders. But we don’t want to bring our ideas to the founders. It needs to be their soul. Right? Their artistic vision.
And so, when we first started, we thought we were going to be doing a lot of B2C, but then, company by company, we just… we just didn’t invest, because it didn’t feel like us… that it would work. And then 4 years later, we woke up and said, wait a minute, we haven’t done any B2C. Why not? Oh, good thing we didn’t, because this whole phase, the window had closed.
The window had closed, but then we hadn’t realized it. The same thing happened with marketplaces. The marketplace would come in, we’re like, man, this just isn’t gonna work, and we wouldn’t invest. So the percentage of deals that we’re doing that are marketplaces has been dropping, not because we want it to be, but because we haven’t found companies that do very well.
That we think will do very well. B2B has been very difficult for the main reason that every node on the network in a B2B marketplace wants the same thing. They’re all fighting over the same thing, and so it’s hard to get everybody to be happy. The same thing, of course, is margin. That’s what everyone’s fighting over. And B2C, it’s easier. It’s like, oh, I want my selection.
I don’t really care if I pay 6 more dollars for those shoes, I just want to make sure I get them in my size. And it’s the quality that I want, it’s the brand I want, whatever. So you can satisfy the consumer, charge them more, and then margin is shared by the marketplace and by the vendor, which that’s what they want, and the consumer gets what they want.
But in B2B, it’s been hard because everyone wants margin. In terms of what marketplaces fascinate me right now. I don’t see one that fascinates me right now. I think it’s emerging. I think, you know, your earlier mentioning of molt book, something, you know, something will probably emerge just spontaneously from all the experiments that are happening. But, you know.
There’s not one that I would point to and say, this is the new paradigm for how marketplaces are gonna work. I don’t see it.

Julia Nimchinski:
Speaking of business models, we are now seemingly pivoting from SaaS and SaaS apocalypse to jobs-to-be-done type of model by agents. essentially charging for the jobs that is done, or, you know, the tokens. What model, business model, resonates the most with you these days?

James Currier NFX:
Yeah. Yeah, I mean, look, I think we figured out about 3 years ago that the big opportunity here was in replacing labor costs. Because if you look at the total IT budgets, software budgets of companies, you know, it’s, you know, maybe 1 tenth labor costs. In almost every company in the world. Today. And so going after those budgets are the better idea. That’s in terms of cost savings.
There’s also the case that what is great is going after revenue generating. So, if we can use AI agents not to just save costs, but to generate new revenues for our customers, I think that will be an even better way to sell in just saving costs and saving time, which I think is fine, but I don’t think it’s as powerful as helping people make money.
I think that’s always going to get people’s attention more. I can increase revenues.

Julia Nimchinski:
info by here. My last question, it’s… just your… your view on the founder team. There’s an article on Infax, this three-person type of futuristic team of founders.

James Currier NFX:
three-person unicorn, yeah.

Julia Nimchinski:
Yes. Would you like to highlight that model more, and do you essentially just seek that type of founding team these days?

James Currier NFX:
Yeah, these days we’re really looking for that, which is decacorns. Not decornes, decathletes, decathletes. So, you know, if you can get 3 decathletes, people who understand code and can manage The, you know, cloud code and the building of the product. Understand marketing and can manage the marketing agents. Understand sales, understand all the different aspects of the business.
You can do a tremendous amount of revenue and growth with just 3 people. And when I say three, you know, you know, after I wrote that article, Sam Altman came out and said there’s going to be a single-person unicorn. I’m like, okay, fine. You’re going one, you know, one is even more extreme than three, and every, you know, whatever. But I still think it’s three, because you need to have friends.
You need to have, you know, compatriots, you need to have a team, you want to have fun building your business, number one. And number two, you want to take a vacation sometime, so you need somebody to mine the store.
And so, we’re looking for small teams that are decathletes and can do everything, all of them can do everything, and then they’re just interchangeable, and they all just manage AI agents, and off they go. And I think that’s, there’s an article called Three-Person Unicorn, which lays out the personalities and the, you know, the skills.
Like, you gotta be good at words, you gotta be good at math, etc, to… in order to cover the bases to really build a unicorn.

Julia Nimchinski:
Huge pleasure featuring you, James, once again. Where should our community go to support you? What’s next for NFX? And and yeah.

James Currier NFX:
Yeah, go to NFX.com, sign up for the newsletter, we’ve got 200,000 people who sign up, we’re the second most popular VC blog after Andreessen Horowitz in the world, because our content is, sent around so much, and… You can follow us on Twitter, and we’ve got YouTube channel with a lot of videos, and then the content, we’re just legendary for our content on NFX.com.

Julia Nimchinski:
have the best content online, that I have to say.