OpenAI’s Land Grab: Are Moats Dead? How Startups Can Still Differentiate - Unsolicited Feedback S3E7

Speaker A: Foreign. Wild west just got a lot wilder and we're riding straight into the melee in this episode. I am Brian Balfour. I am founder and CEO of Reforge. I'm joined by my co host Fareed Mossavat and today we have one of our good friends and AI futurist Aaron White. He was previously CTO at Vendor and now founder of a new AI company, which he reveals in this episode and in this episode we dive deep on moats in the AI era. We then crack open the three questions every builder is secretly sweating right now. 1. Can you actually differentiate when every launch feels like a land grab? 2. Is move fast still an edge or is it just table stakes? And three what happens when OpenAI, Google and Anthropic hoover up all the context you thought was your moat? Expect hot takes on speed as a moat, why Apple's slow and perfect playbook suddenly looks brittle, and how vertical niche founders can still carve out 10 to 100x markets hiding in plain sight. You'll hear the scoop on OpenAI's next platform power move and why it's going to make Facebook's 2007 platform land grab look friendly. A reality check on the Everyone is using AI myth straight from the front lines and a lightning round of optimistic plays. If you're building, investing or just trying to keep your head above the AI tidal wave, keep listening. This convo will arm you with the mindset and the memes you need to survive the quicksand and still grab the crown. Strap in because we're going to get a little bit dangerous before we get to the full episode. In a world where traditional moats evaporate in weeks rather than years, speed has transformed from competitive advantage to absolute baseline requirement. The new bottleneck teams possess unprecedented ability to build and ship, yet they lack the velocity to understand what deserves to be built. The bottleneck has shifted from M execution to comprehension, from shipping code to shipping the right code. Speed isn't just about shipping faster and it is about accelerating your entire learning. Metabolism the problem? It's the feedback fragmentation tax. We collect more feedback from our customers than ever before, but it is scattered across many tools, each with a different owner. This sets up the team for an impossible decision every single time they want to be customer informed. They either got to spend weeks gathering manual feedback or go on a great hunt through all of these systems. The reality is that is such high friction most teams just go with their gut or whatever is in motion. That's where reforge insight analytics comes in. We aggregate all of that customer input automatically. Clean it, structure it, analyze it and put it into the workflows of your team so that they can act on it today. This week we just launched a series of new features helping you absolutely speed up your learning. Metabolism, the ability to drill down to root cause in seconds versus days. The ability to see feedback at a contact account or even segment level. And smart notifications that focus on opportunities to not just updates. Check us out@reforged.com I'd love to give you a full demo of myself. Enjoy the full episode. Everybody feeling energetic?

Speaker B: Yeah. Ready to vibe?

Speaker C: Yeah.

Speaker A: I will start off with a quick story. So I, I got roped into speaking at uh, the AI Engineers World Fair yesterday at this conference.

Speaker C: You did?

Speaker A: Yeah. Um, because they, they, they threw us non engineers a bone and added a product manager track.

Speaker C: Well, I thought product people are engineers now, so it makes sense that this

Speaker A: fair and the, it was interesting. Uh, first of all, it was way bigger than I thought. I was, I was shocked just the amount of people there and the amount of content. But it was interesting. All of the talks were about different technologies, different tactics, different techniques. Ah. The topic I brought to the table was the number one question. How do you differentiate in this competitive environment? Right. Um, and um, I will say it was probably the most sparsely attended talk that I have ever given.

Speaker B: Really.

Speaker A: It was clear they did not want to hear the message that I was sending, which is that this is the most brutally bloody competitive environment that's ever been in tech. And all of your technology does not matter unless you are able to answer the question, like what? Like what do you build and why? And uh, and I thought that was interesting because there was a lot of talk at the conference like engineers should be doing more product work, whatever. You can like define the requirements and all that stuff. But the number one job of the PM is to figure out that question. And so you're not really doing the job of the PM unless you answer that. But there was zero, there was zero interest in listening to me on it.

Speaker C: So you're not doing the job of. You're not doing the job of a good PM unless you can answer that question. There are many are doing the job.

Speaker B: So. Brian, Curious. Sparsely attended but well received by those who did attend or sparsely attended and you bombed. Which one was it?

Speaker A: No, I think it was pretty well received. I mean you like never really know, right? I haven't gotten any feedback or ratings on the presentation yet. I don't know, maybe I bombed, but I thought it was a pretty Spicy talk. So, uh, you know, which is usually received pretty well. But we'll see.

Speaker B: This is one of those things where you say, where it's so clear that, like, while there's a lot of hype and excitement, some of the most important questions are still not well understood yet, if that makes sense.

Speaker C: What are those questions?

Speaker B: Well, I think like on the engineering side, for instance, it's very clear that you can improve your productivity on the work that you are already doing by using AI tools. And I think that that's starting to be pretty broadly adopted. But the question how do you compete in a world where one is the idea of quicksand, that the competitive environment is so uncertain because, you know, things are changing, technology's changing, the rate of adoption is changing so fast that the things we took for granted, which is that if you built a great product that people liked, you'd be able to hold onto it for a while, is sort of falling away very quickly. And that's true at almost every stage of company. And the second is the one Brian, I think calls everyone is doing everything, which is every company is violating the core principle of focus that's been the dominant mantra for a long time. Do a couple things really, really well, and doing product market fit expansion and product suite expansion at, uh, like unprecedented rates. Right. And I think that's still not fully, uh, well understood, maybe with our audience and with people who hear us talk about it every week. But the fact that no one shows up to a talk, that's. How do you differentiate in this new era is sort of an interesting takeaway.

Speaker C: It's funny because the quicksand thing feeds right into that. Like, if you truly feel you're on quicksand, any branch you can grab is worth grabbing. And so you're just going to be flailing, grabbing branches and violating all those principles. And you don't have time to go to Brian's talk to have a thoughtful discussion with yourself or your team about like, well, what is differentiation look like? I, uh, mean, I'll add one particular extra layer on this, which I think is maybe just implicit here. The technology has always been improving exponentially, but we are at the point where the exponential improvement of the technology is at odds with sort of human bandwidth for the first time.

Speaker A: Uh-huh.

Speaker C: And so you actually, it really doesn't feel like you've got time to sit and think because when you watch what the labs come out with, and you watch with some of these, you know, fast growing players are coming out with, you're left Going well. Holy shit. Everything I've been working on for the last, you know, forget eight months, maybe last three months is sort of somewhat invalidated, right? That's how fast it's moving. So I'm a little sympathetic to that, grasping at straws because the only other alternative would be you're a really good futurist and you can see five plus years out and short of that, what do you do?

Speaker B: But even that, what's the answer? Just sit around and wait. You still have to win in the near term. I wanted to do a quick little we're not a news podcast, but I think it's to just illustrate. This would be fun to talk about things I found that have been announced in the last two days. The first is meeting, recording and chatgpt natively, which maybe creates a threat for a whole set of startups, most notably Granola, uh, Otter, meeting recorders, even Gong, et cetera to We've got Google announcing a labs project around talking to experts, right. Which I think is, you know, just yet another thing they're launching and they just got through a whole conference of launching a ton of stuff. Eleven labs just announced a new version of their model. So on the foundational side on voice synthesis and voice creation, some of the examples look orders of magnitude a lot better than previous versions. Cursor just raised $9.9 billion and announced that they raised 500 million in ARR at least. I saw the post about uh, that today.

Speaker A: Yeah, I missed that.

Speaker B: So 500 million ARR for a company that is effectively just a few years old in terms of its like user adoption. Right. And last we have on the data topic we talked about last week, Brian, about the trade off in the Internet. Reddit sued Anthropic for using their content and X slash Twitter just added to their terms of service that you cannot use their developer platform to train foundational models or fine tune models. The war is on in every direction with every single product category and that's just like 48 hours of stuff.

Speaker A: Yeah, yeah.

Speaker C: So what are you going to do? Internalize that or grab it at branches?

Speaker A: Yeah, yeah, exactly. Well, so that is a good transition to the article that I've been thinking about most last week. It's um, from this guy, I believe his name is pronounced Jamin Ball. He's a partner at Altimeter Capital. He writes this great newsletter called Clouded Judgment. He had this intro today, this past week called Moats in the Age of AI. I just want to read the first two paragraphs because I think it helps Set the stage. And I would love for you both to commentate on what parts of it you think are true and maybe disagree with. But he said a lot has been written about moats and software, network effects, switching costs, proprietary data. Everyone wants to believe they're building one. But I've come to believe the idea of a long term moat is mostly a myth, especially in this market. Moats are permanent, they're time bound and at best they function as a bridge. And companies either use that bridge to reach the next defensible position or watch their moat get breached. The real long term moat is just a sequence of smaller moats stacked together. Each one buys time. And what you do with that time, how fast you execute, how quickly you evolve, determines whether you stay ahead. In the age of AI, this is more true than ever. If the moat time window used to be 6 to 12 months, today it's 2 to 3 weeks. Models change, infrastructure shifts, customer needs rewrite themselves in real time. And interestingly, he not was the only one. James Currier from NFX just published an article that was pretty much saying the same thing about speed. Uh, I listened to this podcast from the the founder CEO of Windsurf who shares like the same belief which is essentially saying moats don't exist anymore unless you count speed as moat. What do you agree with, what do you disagree with here and let's dive in.

Speaker C: Yeah, I agree and disagree. So let me back up and then get into the cutoff there. Like on the one hand I actually believe that speed has always been one of the primary differentiators of all startups because the agreed you have no advantages at the start. The only advantage you have is that you're willing to iterate faster than anyone else until you find a signal to pull on. Then the switch changes historically to beating your your one advantage until you've exhausted it and right before you exhaust it, find the next one. And that requires a level of focus on that advantage. And I think you guys have talked about that before, so that's not like new. The Altimeter article here maybe seems like it's saying that second phase just never comes. You're just constantly doing this. And I think that the problem with that is that the faster you're making decisions, the more likely you are to make poor decisions. So I would expect the mimetic frothing of this just to mean that people fall off if speed is their only strategy at some point, because you'll just make a misstep and someone else gets lucky and they don't in the case of Windsurf, I think it's actually like a little telling that Windsurf says this because my belief is that windsurf, GitHub, copilot, cursor, they're not differentiated at all actually. I know there's agentic work that goes on and all that and I'm not trying to disparage. We use their products daily with cursor for sure. There's a lot of good stuff that's in there. But if I think about like ChatGPT first destroyed copywriters and then the copywriting apps and the Grammarly. I mean Grammarly is now a PE firm to acquire other apps based on their brand. They've sort of given up the ghost on like what their primary value was. Like that was the first thing to fall. And I see code increasingly as technical copywriting. So I don't. I think they have to be fast because the, the, the inherent differentiation is just not there because most of the lift is being done by the models. But I, what I think is truer in this world is if speed is the presumed moat and you never find the signal to beat on or the actual moat that you get to like a data mode or a network moat, then it should just. There's no point in investing in almost anybody because people could just make missteps. But I actually think if you've got a huge froth of applications out there all moving fast, evolving wildly, right then the moats that classically mattered I think matter more, which is like who's got the brand, who has the network to reach an audience and who has some sort of compounding data advantage. And in fairness there are very few companies that have that stuff. Like that's actually pretty rare. I think that the problem is people are so used to pitching about their data advantage to sell customers and sell investors and it never was all that real, that's fine. But you're get eaten alive by each other at this point because you're going to be too many apps chasing the same thing because it's too. Because R and D costs are going to zero, right? But go to market costs are where those dollars are going. So having that brand or having that customer lock in based on something that's performance measured, that actually can deliver and a data loop improves it. Humans don't want to evaluate brand new software every day. And I think if agents are evaluating other software, it's got to be based on some hard metrics for them to want to make the switch. You know, in this Future world where agents are deciding your software stack. So it's like I don't, I'm not sure I quite believe this. I actually think it's going to the shore is going to pull out and show us who really had the network data advantage moats for the first time in history because R and D costs are going to zero.

Speaker B: I think, I think calling speed a uh, moat is a mistake because I think everyone knows they need to move fast.

Speaker A: Right.

Speaker B: And I think the overall speed of execution is a rising tide right now. It is moving very, very quickly, probably exponentially. You see it in the speed of product launches, even from large companies. So is it really a mode if everybody's doing it? Is sort of the, the question when I think about what it is, it is almost like a minimum requirement now. And I do think that speed, it's table stakes. I think that speed can help and find the things that people are excited about, find distribution. Right. If we look at the vast majority of AI tools right now, the ones that I am um, at least aware of, I haven't worked inside of these companies but they are primarily driven by really strong word of mouth based on, you know, providing some sort of wow experience in the first 10 to 20 minutes that people get excited about, talk about and share with others. Right. It's the speed at which you get to that that I do think is a really meaningful and important thing. I think in um, the NFX article talks a lot about spe of experimentation which I think is not speed in terms of the way we think about it, which is speed of execution. How quickly can I go from idea to shipped product, but rather what's my clock cycle like, how quickly am I making decisions, getting data, evaluating and iterating on that. That that loop, the clock cycle of that has gone up a lot. Some of that is due to speed of execution of course and the size and interestingness of the things you can build. I do think that's really important but I see it as a table stakes thing. I'm sure there's a whole swath companies not doing it, but the ones that we care about, startups, incumbents that they compete with in general, I mean if you look at OpenAI, I mean you could consider them one of the fastest moving organizations of all time in the software business.

Speaker A: Do you want to talk about OpenAI AI in a minute because I have a pretty dark future picture.

Speaker B: Yeah.

Speaker A: But uh, real quick, I agree largely with you Aaron. I was like trying to break down his pieces of his statement one of the things that he's saying is that moats have never been permanent, that they're always time bound. And that has always been true. Right. That has, that has not changed. But I think maybe one of the things that he's saying here is that the window of uh, that time bound has drastically shrunk, maybe shrunk to a point that it's just completely meaningless. The second thing he's trying to say is because of that you just have to sequence from one thing to the next, which I also agree that's always been true. It's always been about sequencing from layering an X curve. Correct. And then additionally, you know, something he didn't specifically say, but I think is worth also mentioning, is that every moat, even historically, like all the network effects, they have a ceiling or a point of where it has diminishing returns. And some moats have higher ceilings and some moats have lower ceilings. And the, the classic example historically here has been Uber where more drivers led to less wait time, which led to more riders, which led to more drivers. But at some point that hit a ceiling pretty fast because going from three minutes of wait time to two minutes and 55 seconds just really didn't make a difference to consumers and didn't attract uh, more riders. And so I think there's another thing that he didn't say here that has, that I do feel like has shifted, which is that the, the ceilings on certain moats have changed. The ones that used to have high ceilings, some have gotten lower and vice versa. Some that uh, that used to be lower, like data network effects has, has gotten higher. And then finally is like, yeah, speed has always been an advantage in startups. You have to move fast. And the whole point of moving fast is you have to gain escape velocity before you can get surrounded. And so that's not different. But certainly I think the window of time you have to get that escape velocity has dramatically shifted before you get surrounded. And so I don't know the end conclusion here of that. Both like the window and durability and the ceilings have changed. I do agree with that. Do I agree with. The only thing that is left is speed. I don't agree with that. I agree it's critical, it's table stakes. Uh, but I think we're going to start to see some folks emerge here with true moats in this new AI world, which we should talk about in a second. But I don't know. Those were some of my thoughts and reactions breaking that down.

Speaker B: Yeah, I think another just thing I would like to point out is that the things we've considered the most durable moats continue to be as strong or stronger than they've ever been.

Speaker A: Like oh, I agree with that. But we should come back to that.

Speaker B: Some of them.

Speaker A: Yeah, yeah, okay.

Speaker B: I think like direct network effects, people to people network effects products like Facebook, Instagram, WhatsApp, uh, messengers. Messengers are sort of a unique case because they don't have really, really strong global network effects in the same way, but continue to be very, very strong. Facebook's opportunity to leverage their network effect to build even better targeting and advertising, something we talked about a few episodes ago, is probably as strong or stronger than it's ever been. But Google's which is a data network effect. So I think data moats are dying quickly and I think that there are new ones that em. But I don't think, I think there are a couple of companies, the foundational models that have meaningful data network effects. But Google's data, uh, network effect around clickstream, data usage, et cetera, pretty weak right now.

Speaker C: Some are getting obliterated for sure.

Speaker B: Yeah, content marketplaces, great example. Like having more content making it easier to convert, stack overflows and WebMDs, et cetera, like dying very, very quickly, those things. But I do think that some of the more traditional people to people network effects are still very strong and I'm surprised we don't see more AI companies trying to build them.

Speaker A: I have, I don't know why I want to hear conspiracy theory.

Speaker B: I want to hear the conspiracy theory

Speaker C: I can actually break. A year ago I was on here going all crazy about the great cacophony which is just that like as R and D costs go to zero and content generation costs drop, there's going to be a flood of stuff. I think there's actually exactly what's happened. Two implications here. One, the reason Facebook is launching or meta is launching these OpenAI models, I don't think necessarily. Well, I know it's not to do the world good. Sorry to everybody working on that. But if they can, if they can, they can flood the uh, the board with AI generated content and slop. It ought to raise the value of validated person to person connections. And I think that's actually the play. So the networks that can maintain the integrity of who is a person better than others will have some sort of advantage there or really just map the real network on and stay there. Right. Like you, every friend you make in the real world you onload M to Instagram and that's how Facebook knows what the real world actually is. But X is people I'm not meeting in the real world. It's going to end up a lot of AI slop. Here's my counter thought to that though. And I think this one will sound crazy now and then like 10 years, no one's going to think about this because it'll have happened and it won't matter. But I do believe I will be the most interesting friend you've ever made by a mile.

Speaker A: Uh, at some point I say this.

Speaker C: Yeah, and it will erode those network connections as well. They won't matter. Like there's uh, a lot of people like, oh, I want to build AI to talk to, you know, my dead relatives. I'm sitting there going the AI will be more interesting than your dead relatives.

Speaker A: Yeah. Have you seen. Have you seen Tollens? T O L A N S yes.

Speaker C: Yeah, yeah, yeah. That's cool.

Speaker A: Yeah it's. Yeah. See this is what I was gonna say for it. You should check this out is you sign up and you basically talk to your like alien AI friend about anything. And this thing has hit like 12 million ARR in the matter of months. Auto personalizing. I think there's an argument to be said that those people network effects be weaker than we think because the, the content, the entertainment, the interaction that I can get from the AI ends up being more entertaining, more personalized, whatever it is that triggers the dopamine effects which is, I know, sad.

Speaker C: Gotta approaches like a physicist.

Speaker B: Like.

Speaker C: And there was an article that I did not read so this might just be headline propaganda but it was that. And I, but I believe it based on what I've seen with OpenAI versus doctors. It said that that AI some of the latest mod an EQ that is like 50% better than the average person's EQ which I don't think is that hard to believe. And you just play that out and it's like no, I mean we get validated by the AI and yet it can make subtle suggestions that I would never tolerate from my friends that can't possibly soften the blow of critiquing my behavior. But the AI can thread the needle, right? And I'm also more open to telling it more because it's not a person not judgmental. So it knows more about me me. It's a bit weird future. So like I think that's a. I think, you know, maybe Facebook's my conspiracy theory. Facebook's flooding the board to ruin other social networks. But I think they're going to get, you know, orbit nuked. In the end, I think the Google one's really interesting. And it's like you watch what happened with stack overflow, right? Like if you're doing basic knowledge search, not frontier news search, like game over for sure. But you know, some of this other stuff, if you can find a corpus, all that work you put into making that community to generate that content, just gone. Google loses that search. Front door. Oh boy, it's going to be bad, you know, it's going to be really bad. And I, I'll throw one little nuance out there. So we build AI agents daily. What we've seen is like a lot of the models like OpenAI and Anthropics are really good at tool calling. They make good agents. Gemini is terrible at tool calling. In our experience, it's not a good agentic backend, but I think it uh, I think you can understand why. Clearly it's like Google must win the front door, it must win the relationship and it must win the conversational tone. It does not need to win tool call. And in our experience, Gemini has tested the best as being that, like that front door to a chat experience, being able to go deep and understand things and have a conversation, but not being sort of the synthetic backend. So like it kind of paints for me this picture that they realized, I mean, it's probably obvious, but they realized that they can't lose this front door because then that data loop that they have gone, just utterly gone.

Speaker A: Another one. Another historical network effect that I think might be a little bit weaker than we perceive is these cross site network effects in marketplaces between supply and demand. And so it's pretty clear in like labor marketplaces, like the Fivers and the Upworks. Yeah, like those are getting destroyed. But credit to Dan Hockenmeyer, who we mentioned last week on the episode, Chief strategy officer at FAIR, he left a comment on one of my LinkedIn posts that seeded an idea which was actually other cross site network effects of other marketplaces might be weaker as well because those marketplaces technically exist to reduce transaction costs. Whether those that transaction is finding, trying to understand quality, like those types of things. And so I think a good example here is like something like thumbtack, like finding the best plumber in your area. Well, if I can just, I can now probably just plug that into a deep research chat of best plumber in Marin county with some other criteria. Way easier for me to express that than going on Yelp and filtering through all the ads and all the fake reviews, all the Bullshit that I have to put up with there. And it's going to give me a really instant, not instant, it's going to give me a personalized result and do all the extra research that I might need to do manually. And so I think that's another example of where cross site network effect probably weakens as well.

Speaker B: We talked about this a little bit. I think memory as a concept and personalization around who you are, what you care about, what you're interested in is probably one of the meaningful moats that still can exist because you don't want to repeat yourself and you will go to the things that give you the best quality. And the problem for these marketplaces is that they exist for a single problem that you want to solve. Right. I want to find a restaurant, I want to find a, uh, service provider, I want to, I don't know, find answers to code problems. But you can imagine it's still very, very nascent. But the idea that I think the aggregators, who in this case are the foundational models, have the best opportunity to do that work for you because they have context across all of your, uh, context, which is really what the real network effect of Facebook and Google are, is they know who you are across all the properties you visit on the Internet. Now imagine that plus real preferences.

Speaker A: I know not just what you've clicked

Speaker B: on, but what you say and feel.

Speaker A: This is where it gets scary.

Speaker B: This is the dark timeline.

Speaker A: This is where I get depressed. So let me, before I kind of tell why and where I think this is essentially going to go in the next six months is like a little rewind in history. Right? So Facebook launching their Dev platform in 2007. Fareed, I know you were right in the mix on it. Do you remember how many users that they had when they launched that, that platform?

Speaker B: And maybe 50 or 100 million MAUs.

Speaker A: 25 million, right? Not that much.

Speaker B: Yeah. The reason I remember this is because every 10 years I go back to the, uh, Facebook recruiter email I got that said, we're at uh, 50 million MAU now. We'd love to have you join the team. And I said, you're too big for what I'm looking for right now. So I love going back to that just to keep myself humble.

Speaker A: So here, so here's what happened. So Facebook, 25 million users, their network effect was a direct network effect. They needed density of people, right? And at 25 million, in the grand scheme of even the US let alone like international regions, was pretty small. And so they launched the dev platform. And there was this exchange, there was this value exchange which was like, here we created this canvas for you. Put whatever you want in there, games, whatever it is, we'll help you grow it. Whatever you put in that canvas, you can make money off of it. We're fine with that. We just want the advertising on the side rail. Right. That was, that was how it started. And of course instantly they went from 25 to 50 million users in, uh, about four months and two years later they had 10x250 million users. That escape velocity on their network, effect on their specific moat and defensibility. And of course, as anybody who is part of that world, like Fareed and I at the time can painfully remember what happened once they got more confident about their escape velocity, they kept pulling things back. Right. And they were finally like, oh, that money that you're making inside that app, actually, we want a piece of that.

Speaker B: You owe us 30% of it.

Speaker A: Yeah. Until they took it all the way to basically shutting it down and absorbing the highest, like use cases. That was like, that was essentially their decision.

Speaker C: That's going to seem benign compared to what I think OpenAI is going to do.

Speaker A: Exactly. So this is exactly where I'm going. And so the exact thing, same thing is going to play out, but rather than people density, it is about capturing context in memory. And I think what's going to happen here in the next six months, maybe even sooner, is they are going to launch some type of platform and there will be some value exchange for other applications and other businesses to come onto that platform. And it's going to be about more interaction, more capturing of the context into OpenAI's memory systems and that data systems. And as soon as they believe that they have escape velocity on that thing, they are going to flip the value exchange and either tax the whole ecosystem or do exactly what Facebook did, which is essentially shut the platform down and absorb all of the major use cases. And I think that is, uh, not a great ending. Yeah, like for this all.

Speaker B: Yeah, it looks a lot more like the Microsoft model, which is, I think they will absorb the highest value use cases and still have developers because the context is so valuable in other ways. Advertising driving their own usage and revenue, continuing to build the moat. But remember, like, compared to Facebook, which looks like the fastest growing thing on the planet when it was really getting going. Correct. OpenAI, like ChatGPT is orders of magnitude

Speaker A: faster growth than that.

Speaker C: Like way faster. Let me, let me feed into this and give you proof points that this is already happening. So Number one, they just announced, I think it was yesterday, maybe today, I think it was yesterday, that the free tier of ChatGPT will now have memories. Surprise, surprise.

Speaker A: Yeah.

Speaker B: I mean, that would be crazy not to do.

Speaker C: Now their agents are going to have memories built in as well, which gives them developer lock in. Because right now as a developer, it's largely fungible. I'm just sort of chasing some combination of performance and cost. Right. And that doesn't work for them. But here's the, here's what I think the economic physics of OpenAI and the large foundational model companies are. So Google them anthropic, whoever else you want to put in this bucket. We are still on the up part of the S curve of intelligence. And it's a massive capital intensive game that very few people can play. The giants can play at the same time to get breadth of usage. There's this war to bring the costs down as much as they possibly can, including subsidizing them. So I did a presentation for investors recently. The takeaway is smarter than everyone, cheaper than anyone. That's the thing to hold in your head. That's where we're headed, if not already there to some degree. Now you can't play that game forever. If you're selling at a discount, you know, trying to just raise infinite capital from, you know, everybody on the planet, you're going to sort of run out. So what happens, you look at what is the most, the easiest, highest value thing to address with the, the intelligence per token that we're delivering. It's code. That is why they launched Codex. That's why they're acquiring Windsurf. I think the Windsurf founder, I mean, I'm sure he's made a ton of money, but just to deflate the whole like, speed is my competitive advantage thing, it's like, I think you sold out because there is no moat that you can defend against when this is sort of inherent to the models and a few agentic techniques away from making something great. So he made the right move in selling, but I think it's because there was no actual moat. Now granted, selling at that scale without a moat, fine. High fives all around, right?

Speaker A: I'm not, I mean, you know, poor

Speaker C: you, I'm not, but, but you see what I'm saying, like in terms of the physics. So what I think is going to happen here is that OpenAI will keep providing cheaper intelligence than the economics warrant in order to find the next highest value things to absorb. First party. Brian, I don't even Think they need to have an app framework to do this. Like I think it's enough to lock people in with a little bit of memory or just to observe what's passing through the token stream in order to get there.

Speaker B: I am already feeling it. I am already feeling this. I heard from Brian the other day the new cloud model does a bunch of cool stuff really well and I just feel myself being like, you know, I've got a lot of context in my ChatGPT projects right now. I've got some memories, I've got workflows, I've got special instructions, I've got all these different things that make it feel personalized to me. So not just intelligence, but directed and personalized intelligence. And it is driving me away from experimenting with other stuff. For instance, I love Granola. It's an awesome product. I take notes with it on every single meeting that I ever do. It feels like human assisted intelligence. That feels really, really great. But you know what? I don't use the AI chatbot thing that's on the side where you can ask questions to get email follow ups and action items and those kinds of things. Generally I move Those notes into ChatGPT's context and existing projects because I'm working on bigger projects that don't make sense to do in my note taker, I'm working on a new strategy doc or I'm working on uh, you know, presentation or I'm using that as input into some broader like set of stuff that I'm doing and it's sucking up context. Now the tools for this stink, there's no good way to do it. It's a lot of copy paste, it's a lot of mess. But a world where they are collecting that context over time does create a real moat. So I have a hard time coming up with moats for people building on top of these things. But I have a lot of arguments for why the foundational models can build meaningful moats over time and is the net result. There's only two companies share my particular

Speaker C: strategic bent because I'm trying to like practice what I preach but. But I think it's like as good as granola is, you've said basically I value the UX and the utility. It is not a platform for me and that is unstable ground for a piece of software to be on. Right. So you know, if once OpenAI launches their meeting note taker and it sounds like they have it's going to be apocalypse, you know now and they hoover up more content, more context, not just for the individual but they can start parceling out between individuals what the shared context is. We said. I think I said this a year ago as well. I think people in a good position that haven't um, like raised their like Dragon mode yet, like Apple ah has not gone founder mode on this just yet. You know, if you own the device, you see everything the user sees. It's funny, Limitless switched from that rewind app to the pendant. But the rewind thing actually made a lot of sense to me and I think the OSS will start doing this at some point. Just be like look, whatever you see, we see so we can be better for you. That's going to be the ultimate integration point. I think that's why OpenAI has their operator. They can be there trying to passively absorb this now and it's why they're trying to build their own device. It's not just and browser probably because they need all the content and device. So I think like hey, knock knock, you know, Google, Microsoft, Apple, wake the hell up. Like it's a bit of a privacy nightmare but that's where this is headed for sure.

Speaker B: So just to get back to the speed argument is probably most acutely obvious Looking at Apple versus say Google, Microsoft, Facebook, et cetera, the other faang companies. Apple's strategy, which is to be last mover worked really awesome for a certain category of stuff that they have dominated for almost 20 years. We're almost 20 years into the iPhone era. That has worked incredibly well for them. I think it's a huge liability for them right now and you just see it, they're trying to move fast by announcing stuff and then never shipping it. Siri is not a useful product at this point in time. There's a lot of ways they could move fast. But I think their fear of being beholden to some of the foundational model companies probably is a piece of the puzzle here, uh, which they have traditionally owned. All of these things plus culture of slow deliberate moves is really hurting them. So this is again speed is table stakes and the companies that have designed themselves around not being fast struggle. I think the same is true for Anthropic versus OpenAI. Anthropic is deliberate. It's part of their core values that they have safety and you know they're going to release things later as a big piece of what they do and I think it hurts them especially on consumer adoption.

Speaker A: Yeah, just side thought here. Basically the move would be Apple should buy Anthropic. Right? That's the.

Speaker C: It feels almost Inevitable AWS or Apple buys Anthropic.

Speaker A: Yeah, um, that's.

Speaker C: And my anthropic equity is ready for that. So, you know.

Speaker A: Wait, what, how do you. Anthropic equity?

Speaker C: I didn't get in early, guys. This is like. So I don't know if you've seen behind the scenes. Every deal now is super syndicated to everybody in their own. Oh yeah, there's no alpha anymore except like the seed stage or the Series A. So if you're an investor, you just get to see every, every series C and beyond and every secondary round. You can throw dollars in very expensive deals. But there's certain ones like Anthropic that I think will be like at least, uh, you know, some sort of multiple on.

Speaker A: So if you are not OpenAI or one of these major players and you are a company competing in this environment and this, this feels like the inevitable vector that we end up down, like what do you do? What do you. How do you behave?

Speaker B: You know, Aaron, you're the, you're one of the folks building in this space right now. What's your approach?

Speaker C: So I'll tip my hand as to what we're building if it's helpful because that might provide some context. Again, I'm going to die on this hill. Hopefully it's the right hill. So we're building a platform where any domain expert, someone with any specialty knowledge can come and talk to talk to our platform and it will build them an AI business soup to nuts. Landing page monetization, customer management and the actual agent or app. And you do not have to be technical for this. If AI can write code, you definitely don't need to be technical in the future. That's an implementation detail of the highest form now. So I share this because if you think about like where every vertical AI SaaS comes from, it comes from a domain expert being interviewed by product who is interviewed by an AI, you know, pipeline team that then hands it off to the SAS infra team. SAS infra is meaningless these days. I think you can automate a bunch of the AI work. We're learning the patterns and if you can automate a product person. Sorry, Brian, then you can get this. You just need the domain expert at the top. And the reason I share this is the thing that's happening in the agentic field right now. And I don't know what the window on this is. Just to be clear, I'm painting for you what I think is the longest window I can find to get an AI agent working well does require A bit of a loop of experience and does require the type of knowledge you only learn through the slow school of hard knocks. If it's deducible.

Speaker B: Specific knowledge is what I think. Things you can only learn by doing. Doing.

Speaker C: If it's common knowledge or it's logically deducible, the foundational model is going to eat your lunch if it's something you kind of have to live and breathe to learn the nuances of someone. Taking the time to encode that into an agent and then increasing its efficacy by collecting those data loops is creating a mini M moat. And as long as those moats are like relatively mini enough, I think you've got a somewhat time horizon defensible thing. If it's something like how to do good code I don't think it's defensible at all because it's so massive. It's ah, at such interesting scale that the head of that specific knowledge curve is going to be eaten away. Doctors, copywriters, like all that stuff. Yeah.

Speaker B: Competition in horizontal or large M markets is going to be too much. It's just going to be too much noise I think is the thing you're saying. Basically.

Speaker C: Yeah. So what I'm trying to do is aggregate up the long tail.

Speaker A: Mhm.

Speaker C: And I think there's some defensibility both for everybody who chooses to build with us, but also in aggregate for us. A lot of problems with the strategy clearly. Right. Like could probably rattle them off right now. But I think it's like the specific knowledge is the one thing that I can get behind right now that I think is immune to uh, just rapid absorption. The other one will be any sort of person to person network effects or durable brand trust. But we've already talked about why some of these things are not as long term real as we thought. But as the cacophony goes up, compounding advantage that people have figured out, go to market or just own an asset there that allows for that. If you could figure that out today, run to it as fast as you can.

Speaker B: So one interesting takeaway from that is I do think uh, I think people in the valley have been calling this taste a lot recently. I feel like I've seen a lot of thought pieces on like how important.

Speaker A: Rick Rubens.

Speaker C: Yeah, yeah.

Speaker B: And I think taste is really in some respects uh, a proxy for specific knowledge. You know enough to make smart decisions that are novel or interesting because of your depth of understanding of a specific problem. I saw a tweet that was like being good as a PM means Being able to look at a landing page and know what its conversion rate will be.

Speaker A: There's Nikita beer. Yeah, Nikita beer.

Speaker C: Yeah.

Speaker B: And in some respects that is the knowledge that you develop after having run a billion things.

Speaker C: You just know what works agent to do that. Let's go do that.

Speaker A: Yeah. Ah, exactly.

Speaker B: So there's a world of that kind of stuff. And in a lot of ways it's the kind of work that I do is like, hey, let's short circuit a bunch of decisions. Well, let's drive speed by having good taste, by having good specific knowledge, by having good decision making. So I do think speed of decision making is like the most important thing. And I think we call that taste, we call it knowledge. And I do think there's a lot of opportunity. Ignore the like normal venture capital path of like building really meaningful businesses in a lot of these corners of the world without needing all of the expertise you used to need to do. Like, if the cost to entry is much lower, the number of interesting businesses should grow. Now I think moats, I don't know, like 10 other people who are as smart as you can build them too. But maybe that's okay if you don't raise $100 million to build.

Speaker C: I think that that's, that's actually a very important term which is there might be a lot of implicit here, I believe this, but like a lot of interesting size successes that don't require venture scale dollars. But actually to your point, on cycle time, I think it's, it's actually really important because I think you've identified three components. One is specific knowledge allows you to move faster because it gives you the confidence. Two is tooling, like your awareness of what's possible allows you to move faster if you lean into the future. And the third one that I think is implicit in a few of the examples you've given is operational excellence. Like that culture that allows you to do that is actually a hard thing to generate because it's against a lot of human instincts. So the Bezos quote's so good about one way versus two way doors because most humans default to thinking every decision is a one way door. And that's actually not true. And so if you can fight your instinct, you can move faster. And that's a sign of operational excellence because it means a lot of internal trust both in the process and each other. And that's hard. But that gets easier as organizations get smaller because there's no one to justify anything for.

Speaker B: Brian, how about you? How are you thinking about speed of These topics right now, like, what is the advice you would give yourself or other people building who aren't, you know, the two biggest companies?

Speaker A: You know, the talk that I gave at the conference borrowed some of the stuff that Ravi and I did for the AI, uh, strategy, strategy program, which was talking about his concept of finding differentiation is about treating AI like assembling Lego pieces. Right. Uh, in this, in the sense that you don't have to build a whole new model and on one end of the spectrum and on the other end you can't copy and paste kind of the chatbots and expect differentiation. There's something in the middle that combines some form of data, ah, some form of off the shelf AI components and some form of your functionality together in a unique way to find a seam of differentiation. And granola was the common example there. Right. Like they entered a market that was just all out competitive, but they combined a few things that were very unique. They um, combined not just the transcription but with the person's notes to generate a unique output. That unique output started to accrue, ah, into the library of their notes which they could start to do additional things on top of like chat and the projects and the cross workspace functionality. And they blended that with very specific functionality of doing the Mac app as well as the tool calling into your calendar to grab the metadata and those pieces to create a delightful habitual experience. They just used all the off the shelf AI and so that found them a seam in a very competitive market. They so far have sequenced those Legos into another set of Legos. But now the question is, are they able to still keep doing that? I think that is a question for, uh, a lot of teams internally. I think it's um, trying to redefine what we think our own internal definitions of speed. I think that's probably the hardest part is, uh, everybody has to take a mentality of playing bigger because what you can do is a lot bigger. And look, I've been in the industry for like 20 years now, so that's 20 years of habits and thinking about what is possible with the people and all this stuff that I have to somehow short circuit and rewire on every single decision that we make. And then I have to get everybody else in the organization the same way because every time I'm saying let's do this, of course the natural reaction is, oh, that feels crazy. No, no, no, like let's reconsider the new environment. And so, yeah, I think that's just, uh, James had talked about in his, in his in his article that like speed is a lot of it is like a mindset shift and, and it is and it's. It has taken even though I know all this stuff it take. It takes a lot of proactive effort to shift the mindset and redefine your expectations of speed and what you're able to do.

Speaker C: Your years of accumulated knowledge make you a potent strategist and expert, but also ossified in the way you work.

Speaker A: Totally.

Speaker C: I mean we're feeling that we're building AI software with a bunch of AI interested engineers. But are we using it enough? Like I know some of the you know the kids who are running laps around all of us are doing like uh, probably not one. One interesting exercise that I think this is maybe not for the whole org but for engineering teams out there. And if you're already doing this and I sound like a dinosaur, good on you. And if you're a dinosaur listening in like it's. It's useful. Uh, choose choose some period of time, a couple of days a week by Devin, get it set up ahead of time and then force all engineers to only interact by directing Devin, you may not touch your code editor. Find opportunities like that to force people into a. Yeah.

Speaker B: One other thought that came to mind Brian, as you were talking is how do you approach big enough ideas that I don't want to say small, but small enough that they are out of the wake of the major players. And the way I think about this is looking at Microsoft in the early days of the PC is they absorb the largest, most obvious horizontal end software use cases into their ecosystem as fast as they possibly could. Like that basically is the Office suite. Things like Outlook, file sharing, you know, um, I'm trying to think browser like those are so big that they just get absorbed by the platform. But there's a huge Microsoft to this day PC software ecosystem. It's huge. It's way bigger than Microsoft collectively. But each of those its own things I think vertical solutions have have a unique subset or segment that they're that are. That they're interesting to et cetera which is where I think it's really hard because you want to go for the biggest opportunities available. But if you think about what LLMs are really good at like code writing, meeting context like doc creation like these things are right in the wake of the no totally container ship size thing moving at 100 miles.

Speaker A: I think the definition of like what is unique and differentiated as is completely shifted as well. Because to our earlier conversation it's very clear that both OpenAI and Anthropic are going after hoovering up as much of business and in the case of OpenAI consumer data as possible, whether that's through integrations or through their own native functionality or through this eventual app app platform that I think they will likely launch. And it's clear like the things that they probably don't touch are things like legal and health tech. But anything in the general business environment feels fair game and open territory for them to, for them to hoover up so that just that bar of what is different and, and what is protected is fundamentally different.

Speaker B: No, my m argument is every market is at least 10x bigger than you think it is. Well, that's the future. So it's okay. Like, does that make sense? Because like I think we saw this in cloud. It's like what looked like small niche markets were actually massive. I think they're even more massive than that.

Speaker C: So are you bullish for a Gleam?

Speaker A: No.

Speaker C: Right. No, I think too close, darling of the enterprise AI, you know, recent iteration. And yet with these launches of these connectors, it seems like it's going to be a hard one. Maybe the market's 10x bigger than I think it is, as always. And um, you know, quality and sales matter here. And that may be true, but it's. The moats to Brian's earlier point are getting obliterated quickly. So you really, I think you need a deeper theoretical framework if you're trying to be big. Because one last thing, seven to 10 years to IPO feels like a liability these days, like a real liability. You know, there'd be companies that get to hundreds of million. Take Gleam as an example. I don't know, they got started exactly. But like you just, you're not going to get liquidity out fast enough before the market shifts out from under you yet again. And I think like Microsoft's ecosystem actually gleams fundamental problem potentially again, very successful company. I'm not trying to disparage it and the founder's awesome, the product's awesome. But like you, you know, all that Microsoft ecosystem is built to support humans doing some work at the tip of a spear. But if that is being, you know, ah, end run around right up there, that all is going to collapse ridiculously fast because it didn't have a theoretical safety.

Speaker A: Yeah, yeah. All right. I think I feel like since we're also running out of time here, we have to end on some positive thought or so maybe we just do a, uh, quick round the horn and try to leave this conversation with a Little bit of uh, optimistic energy here.

Speaker C: Computronium supply chain. Go dig up minerals, make chips, power them and put them in robots. And if you're doing that, you're probably actually like the software industry as we know it has taught us a lot. It's actually applicable in more places. Go find your way into that supply chain. That's be my positive optimistic note. I think there's a ton of growth there because we already know we're going to be constrained on throughput with this stuff.

Speaker B: My mine is that thought around markets are 10 to 100x bigger than you actually think they are. Your ability to, if you know how to go to market and have specific knowledge in a specific in an area, you can build organizations and companies that are much smaller and much more efficient than they ever have been. I think the opportunity for what look like what used to be called small cute lifestyle businesses are actually 10 to 100x bigger than you think. And so I think if you can, if you know what those things are and you can and you can build in them, it'll be a while before somebody, before the big players come for you. The danger there is all the other little guys, but that's a much easier game. That's the normal startup game.

Speaker A: Uh, yeah. I'll leave on a note. We're still early in the game. The three of us and probably all of our listeners live in a LinkedIn Twitter bubble where it feels like it's game over. Everybody's adopted everything at this point. And I can tell you like we have been running interviews nonstop with people on the front lines adopting these tools. And one of the main questions we ask is like, okay, tell us about the adoption and the rest of your organization and your friends in the network around you. I mean, outside of the typical general usage of using a Gemini here and there or a chatgpt there, the story we find all the time is there's a couple people in the organization really using it in depth. Everybody else is not. So it's crazy to think about we've seen this much impact and chaos given that we're maybe just exiting the early adopter phase of, of the curve. But the game is still early.

Speaker C: I like that. So we should all also invest in the foundational models series, you know, plus rounds, apparently.

Speaker A: I get it. I need to get some of your anthropic actions.

Speaker B: They all need to just be public so we can all bet on it. Well buy Nvidia maybe. Is the takeaway there?

Speaker A: Yeah, that's fair enough. All right, Aaron, thank you, man. This was fun. Appreciate you joining.

Speaker C: Thanks, guys.

Speaker A: All right, everybody, that's a wrap. I'm Brian Belfor, founder and CEO of Reforge and the host of Unsolicited Feedback. Today we tore through a few key topics. The myth of moats, why speed is table stakes. Context is king. The escape velocity window keeps shrinking as well as platform power plays how the most likely next move from OpenAI could make Facebook's 2000 land grab look absolutely polite. If you're feeling both thrilled and mildly terrified, you're in the right place with all of us. When the dust settles, the teams that win will not be the ones with the snazziest dashboards. They will be the ones who really own the why behind every customer shakeup and understand their unmet customer needs. And that is exactly what Reforge Insight analytics does and helps you do faster. It aggregates all of your customer input for all the sources. It combines it with your quantitative data, helps you analyze it automatically with AI so that you can turn that chaos into laser focused revenue producing insights incredibly fast. Let me give you a spin on it. Let the AI do the archaeology and ship product that actually moves the needle. Visit Reforged.com and we'll give you a free trial. Thanks everybody. See you next time.