Lesson 6: Fixing The Growth Problem Using AI Paywalls

Transcript

(0:03 - 0:12)

Hi everybody, welcome back. Today we've got another one for you here. I'm gonna go ahead and share my screen.

(0:14 - 0:31)

Okay, so this is a follow-up to the product growth strategy for Project Bitcoin. This one we're taking a little bit different tact. It's the growth problem and specifically AI paywalls.

(0:31 - 0:47)

So this is, we're gonna start moving from strategy and start moving into product. And as we often do, we will work through a number of topics here. So here's the overall table of contents.

(0:47 - 0:59)

I got about a hundred slides to get through. I don't think it should take that long. It sounds like a lot, but much of it is in storyboard format and that's just to help tell the story a little bit better.

(0:59 - 1:07)

So we'll work through these step-by-step. The problem piece, we've got a couple of subsections. So let's go ahead and make a start.

(1:08 - 1:22)

All right, so overview section, let's define some terms and look at the current state of affairs. Let's start on the right side first. So state of affairs, the emerging tech stack, spatial computing, AI, Bitcoin.

(1:22 - 1:48)

Spatial computing right now where we sit in mid 2025, Apple has their device out, Meta has their devices out, Snap has a device. We're in this weird in-between phase where we haven't crossed the chasm to mass consumer adoption for the new interface layer. We've got really high performance noggin goggles, if you will, and then also sunglasses without the high performance and capability set.

(1:48 - 2:16)

And so over time, we'll start to see those compressed. AI, on the money being put into this space, new capabilities coming on the weekly, kind of taking the world by storm over the last few years, especially with reasoning and agentic workflows. And then Bitcoin, as we mentioned last time, it's reached a really interesting place from an investment perspective, but really our definition of this is, it's a way to store oil on the internet.

(2:17 - 2:25)

It's weird, right? But factually accurate. So then on the definition side, let's cover these things quickly. AI agent, won't go too deep into that.

(2:25 - 2:31)

You can read about that online. Atomic decision units, these are our nomenclature. You won't hear that elsewhere.

(2:32 - 2:49)

We'll get into that in the sixth element of this. Bitcoin Lightning is a layer two protocol on top of Bitcoin that helps you do a high transaction throughput, low cost transactions. Can't do that on Visa, MasterCard, ACH, Wire.

(2:50 - 3:05)

Can't do it with Stripe. So it's one of those things, if you want to do sub penny transaction, you're going to have to move into the world of digital currency and what bigger, stronger, badder network than Bitcoin. Decision meter, we'll get into that as well.

(3:05 - 3:17)

This one just touching on it, but product recommendations is a feature set that we'll take care of. We'll get into that later. And then MCP servers also taken the AI world by storm.

(3:17 - 3:32)

Essentially think about this as a way to standardize APIs from different companies. That way AI agents can talk to, read from, write to any consumer app or enterprise app. So, okay, let's talk about the problem.

(3:32 - 3:44)

Shark week right now, so why not? Sharks, problem, solution, who knows? All right, so I'm going to walk you through a couple of things here. So humans, we're a big deal on this planet. We like to think so.

(3:45 - 3:50)

What do we want? We want to make less decisions. We want more time. We want things to be cheaper.

(3:50 - 4:09)

We want to make more money. There's a ton of us on the internet these days and we seem to be pushing around a lot of data and information and knowledge. The other side of this, where I've spent quite a bit of time, enterprise SaaS applications, what do they want? What do investors want? They really want to monetize their data.

(4:09 - 4:20)

They want to reduce costs. They want to expand multiples. There's a lot of these out in the world, probably more than you would think, about 30,000 SaaS companies, 50 million customers of these companies.

(4:21 - 4:46)

Think about small businesses, individuals, and over half a billion users, which is pretty wild. Okay, so then what about the intersection of this? This is the problem space, right? So AI agents, can we somehow magically connect these two things together? The answer is yes, but how? Sorry, my dog's going crazy. Nash, come on, bud, calm down.

(4:46 - 5:06)

So this is really the opportunity space that we're going to dig into today. So, all right, as we dig into this problem, let's talk about the enterprise problem a little bit more deeply. So enterprise SaaS on the right side here, big thing, we want to increase profitable growth, but let's break this down.

(5:06 - 5:26)

So what is an enterprise SaaS application, software as a service? Well, they have a user interface. Most humans understand what this is. A lot of customer development and interviews over the years have led me to understand the frustration and pain of the users.

(5:27 - 5:37)

Too many tabs, too many apps, too many buttons, too many pages, clicks, workflows, yada, yada. It's getting pretty bloated. Application logic is difficult to understand inside the company.

(5:38 - 5:48)

They may not have been documented. A number of different engineers or pods all around the world have worked on it. Different code bases have been brought together from acquisitions or new builds.

(5:49 - 6:02)

There's frameworks that have been included to accelerate the product development process, but they haven't been updated. There's security patches that need to take place. Underlying languages need upgrades, but haven't done it.

(6:02 - 6:16)

So there's a lot going on. And then on the database and infrastructure side, we may have a mix of cloud environments across Amazon, Microsoft, Google. We may have some on-prem stuff, some colo stuff.

(6:16 - 6:33)

We may start using model training environments and spinning those up. So it's getting pretty complex. And then the APIs that allow other companies to integrate with them or customers, a lot of times there may be private ones, some public.

(6:33 - 6:48)

They may not be documented very well, Swagger or otherwise in some cases, which is great. They may have different sun setting capabilities, different versioning, usage may be unknown on these. And every company's API set is different.

(6:49 - 7:05)

That's a fact. So it makes it really challenging, even though we have this restful capability set, the endpoints are different, the authentication mechanisms are different. So it's a mess.

(7:05 - 7:29)

This is difficult. This is why you have teams of really experienced, sophisticated people just trying to wrap their arms around this. But it presents a problem in this new economy with these things called AI agents and how are they supposed to interact with these tools? So AI agents, you write a prompt, hey, go do this thing for me.

(7:29 - 7:59)

It's gonna attempt to connect to another thing. It's a tool call in technical parlance, but I could call the tool. I could call Salesforce, but if the AI company isn't integrated with Salesforce, how are you gonna get those things to talk to one another? And then aside from that, there's a discovery problem, which is like, how does my enterprise SaaS company get known by these AI agents, especially if it's new, right? So that's a challenge.

(7:59 - 8:27)

So solution, and we'll get into this more a little bit later, but essentially we need some way to like change this blocker into like a gateway and start to allow access that's authenticated, secure, and monetizable. But before we get into that, how big of a problem is this, right? So I mentioned earlier, it's like, this is sort of the scale. Well, if you multiply these together, what's the size of this? It's 10 to the 21st power.

(8:27 - 8:32)

Now, I don't know. I'm a math guy, but I don't know what that means. There's a word for it.

(8:32 - 8:45)

It's called Zeta scale. So we ended up taking a look, and this was interesting. So what are things that Zeta scale? Water molecules, so one drop of water is a Zeta scale H2O molecules.

(8:46 - 8:59)

So grains of sand on earth is at the same scale. The width of the Milky Way galaxy, the amount of data created every day, give or take. The Bitcoin network, this one surprised us.

(8:59 - 9:34)

I was like, wait, is that right? And like, yeah, a Zeta hash computational power every second, right? And then enterprise SaaS, the number of essentially human interactions across enterprise SaaS products. So what's interesting to me is that we're talking about the same scale. So like, should we wanna go down this path of like opening up that gateway between the AI agents and these applications, consumer or enterprise? You need some other type of network that has the scale and power to like sit in there.

(9:34 - 9:39)

And so it's already there. Like you don't have to go create anything. So that's great.

(9:40 - 9:45)

But then here's the challenge. That's just human use. That's not machine use.

(9:46 - 10:04)

So machine use is gonna dwarf human use, right? So then what do we do? So let's dig into this growth problem here. So I'm gonna dig into this a little bit deeper. We're gonna go through the whole thing, zoom all the way out to like internet writ large, right? Humans, applications, machines.

(10:04 - 10:17)

So let's go back to like fundamental wants and needs and desires. So humans, what do I want? I want more time, right? What do these companies want, these consumer brands? They want more money. They wanna grow the business, makes sense.

(10:19 - 10:32)

And how do they do that? Well, they need to get more money from the consumers that buy things from them, subscriptions, products, services, you name it. But there's a problem. Consumers are the source of value, but their earnings are being eroded.

(10:33 - 10:57)

Their money is being deflated and their earnings are not keeping up with the increase in prices. So this equation isn't gonna balance out. It's getting worse, right? So we need actually the primary growth problem is with the consumer not having enough cash to go around and not having enough time to earn more cash or even do the things they want.

(10:57 - 11:15)

So that is the essential blocker that we need to focus on here. So yeah, need more time, need more cash if we wanna make this thing as about as basic as we can get. But therein lies the challenge, right? Like how do we jumpstart this economy, this global economy, if you will.

(11:16 - 11:34)

And so let's talk through this classic growth loop, call this before the rise of emerging tech, AI agents, Bitcoin, et cetera. So consumers pay to use these consumer brands. Consumer brands pay to use this enterprise software to make them more efficient and manage all their data and supply chains and et cetera.

(11:35 - 11:49)

And then this enterprise software companies like pay consumers as workers to like build the software. And so you get this nice loop. Now, granted I've simplified quite a number of things to show this, but it's interesting.

(11:49 - 11:59)

It's nice. It's neat, right? And see that flywheel happening. Okay, so disruption, AI agents are here.

(12:00 - 12:21)

Emerging tech is here. What do we do? Like, do we control the robots? Like, what do we do? We like this stuff here. Okay, so we're back to this problem but we need to introduce something new, right? So we've got these enterprise AI agents that are starting to get built and they need to talk to enterprise software and the consumer brands need to build them out.

(12:21 - 12:32)

So in that case, we're not gonna pay to use the enterprise software. We're gonna pay to use the enterprise agents that pay for the enterprise software. Okay, got it.

(12:32 - 12:53)

And then the enterprise software is gonna pay to build out these enterprise agents. All right, so then if I pay the software to build agents then I don't need the humans to build it, right? Like the AI is doing it for me. So then the money to the consumers go away and then consumer software even pays to use enterprise software.

(12:53 - 13:07)

But then consumer brands, like they need to get consumers so they're paying to advertise on consumer software. This is like your social networks, your Googles, right? And consumers wanna use this for free. That's where they spend their free time these days.

(13:07 - 13:19)

But consumers also have AI agents like Chad GPT, Gemini, Anthropic. And so they're gonna have to pay to use those. So now they're spending money, they're not earning any.

(13:20 - 13:33)

So is there a possibility where consumers would get paid to have them open up some of their data possibly? And then, so it's like, okay, this starts to get confusing. So let's clean this up. And we start to get a bunch of lines.

(13:34 - 13:55)

And basically we're back to this whole thing, which is like, how do you even keep all of this in your head? There's obviously a lot more complexity here. We can't manage this entire ecosystem, or can we? I don't know, let's find out. So what we really need, what we're really missing is a big idea.

(13:56 - 14:07)

So let's go back to our idea here is consumers. They have AI agents, right? Let's go through step-by-step. How should this work? Well, consumers will pay to use AI agents, we're seeing that.

(14:07 - 14:27)

Pay 20 bucks a month to use Chad GPT, ask questions. Now they have agents, they can complete tasks, small ones at first, but their whole definition and agreement with Microsoft is they have to reach AGI and $100 billion of economically valuable tasks. So they are going down this path.

(14:28 - 14:43)

We also have these things called enterprise AI agents, not just consumer agents. We can't assume that they're gonna be one and the same. They're very different needs, capabilities, security protocols, privacy elements, et cetera, scale.

(14:45 - 15:09)

So perhaps consumer AI agents actually pay to use enterprise AI agents, right? To do the things that you want them to do, like more pro-level capabilities behind the scenes. The consumer agent is all about the user experience, simple, easy, fast. The enterprise agent is all about capabilities, right? Deep complexity.

(15:10 - 15:47)

And then you've got your companies and your enterprise SaaS stuff of old, which is like paying to access these tools to get data, workflow, services, products, you name it. So then potentially we could create this loop going back the other way where the agents get paid for like doing tasks, right? And consumer AI agents get paid for passing along workflows and consumers could end up getting paid to help train some of these AI agents. Like here's a daily bounty for your data, your workflows, your decisions, your reviews, tasks, et cetera.

(15:48 - 15:53)

So then you've got a loop again. And so that's nice. And then consumers have some free time.

(15:53 - 16:04)

Again, they can hang out in consumer apps. Maybe they end up wanting to pay for the best quality stuff like they do for streaming subscriptions. So not unheard of.

(16:05 - 16:28)

And so the big idea here is that it requires paywalls everywhere, right? So, but the thing about these paywalls is at least in some of these, like between these agents, these are gonna be like microscopic transactions. It's not gonna be like a hundred bucks a month. So you need some element to manage that.

(16:28 - 16:43)

Okay. So now we're into this potential solutioning here, which is you can start with the same place. It's similar concept overall, right? So you've got your prompts, your payments, you have guardrails outside the agents.

(16:43 - 16:53)

They have some kind of wallet, ideally on Bitcoin. They make tool calls. They request this in Bitcoin.

(16:54 - 17:28)

In front of that is an MCP server, which standardizes the API access. And then I'll work through this workflow quickly, but essentially the way that this works, it's kind of protocol-y, is you need to like request an API, check for funds, have an invoice, make the payment, then get access to the enterprise SaaS data or consumer app data, make sure the payment has gone through, complete the response, move it back through the agent, and then ultimately complete it for the human. Okay.

(17:28 - 17:48)

So that's it. So then if you remember from where we started, really this piece in between is where this paywall will be. It's like between AI agents and these applications as one potential future outcome, right? This is likely how things will play out.

(17:48 - 18:09)

You need these AI agents accessing systems that have been built, right? We stand on the shoulders of giants. So that's really it. And enterprise SaaS, consumer software, basically just agents and software talking to one another and paying for access and data and workflow, right? And so here's some of the things that would drive growth.

(18:10 - 18:26)

It's better task handling, consistent decisions, private workflows and data. You can monetize that as enterprise SaaS. It's a new business model for them, opens up new distribution and discovery channels, and you can start to capture some of this AI tailwind.

(18:27 - 18:52)

So here's a little bit more detail on each one of these things, and we'll publish this as well, this deck you can comment on Google Slides. And so here it is in the middle is like, you really need this number three piece to enable this. And so it's not really been built out as a service, as its own application, that's easy to plug in today.

(18:52 - 19:30)

So you're telling me that 30,000 companies are all gonna like custom build this themselves? Like that just seems like an entire waste of time, effort, energy when they've got to build their own application. It's like, why would they do all of this over and over again? So it's a perfect place for like a product to actually be beneficial to the ecosystem, to the community, to individuals, to companies. So then one more thing here, a key piece of this in this tool calling workflow, which even OpenAI is kind of glossed over.

(19:30 - 19:40)

It's like, oh, the model just calls a tool. How do you know which one it calls? We don't know, makes decisions. That's what people are worried about when they're talking about this AI safety thing.

(19:41 - 20:09)

It's like, they're really worried about the decisions they make. Like, are you gonna push the button when I don't want you to push the button? And then what happens when you push the button? Nothing good. So the real challenge in all of this then is how do we ensure AI makes decisions we can trust when there's this hallucination issue? And so really what you're talking about is you need some kind of like deterministic decision engine, which basically says, I give you information.

(20:10 - 20:27)

If I give you that same information, you give me the same decision, right? Over time, that information may change, which means your decision may change, but given the same information, your decision shouldn't change. Like, just because it's like the sky is blue today, I'm not gonna take an umbrella. It's like, oh, it's raining.

(20:27 - 20:48)

I'm gonna take an umbrella. So you gotta just, how do you program that into a model? Well, you need deterministic software or much smarter software than LLMs, like biologic intelligence. So really what we need is a way to inspect and edit the decision-making process and ensure it's maintaining certain guardrails.

(20:48 - 21:02)

And so we use this shape very precisely. This honeycomb is very strong in nature, but it's actually because we spent a lot of time on this. There's six inputs, we believe.

(21:03 - 21:23)

Now, you may say there's like one or two more, maybe a little bit less, but we think these are the big ones. And then those six inputs get you one output, right? Okay, so how do you programmatize an AI agent's decision engine? Well, here's the various inputs. You have to outline the goal.

(21:23 - 21:43)

So like, what's your goal? What are you trying to achieve? What's the context? That seems to be like 90% of the conversation with AI has been talking about context, but we haven't talked about these other five things. What are the constraints that you're under? Clearly, like you don't have $100 trillion to spend. You have $300 to spend.

(21:43 - 22:12)

So your options, therefore, may be different. So what are the options that we're working with? What are the signals either in real time or not in real time? And then what are the values that you or your company has? Is it just to make money? Maybe it's to make people healthier, right? So all of these things get taken into consideration. And then we create a decision score.

(22:12 - 22:22)

It's formulaic. What's our formula? Well, we spent a lot of time on that, but essentially that gives you one output. So we call this essentially the atomic decision unit.

(22:22 - 22:32)

It's one decision unit. So the atom of that, right? So these six inputs, one output. And then here's the beauty of this is much like humanity.

(22:33 - 22:55)

One decision begets another decision. So you can start to chain these things together. Like one decision output, maybe the input to another decision, right? So given I bought this house and the dimensions of the room are 20 by 10 feet, what couch will fit in there? That's a toy example, but an example nonetheless.

(22:57 - 23:10)

So obviously we could chain these things together and you can go in multiple different directions, up, down, left, right. There's negative space in there. But essentially you can see how some of these things can build up over time.

(23:11 - 23:29)

Decisions can improve reinforcement learning if you're in the community. And so we can start to create a foundational platform upon which we can start to build a trusted AI agent. E, trust, because we trust that you're making decisions.

(23:29 - 23:38)

Given you make a decision, I can pull it up in my interface. I can inspect that decision. I can inspect the goal, the signal, the value.

(23:38 - 23:47)

I can inspect that element and that formula. And I can say, here's exactly why we made this decision. Given the same information, we would make the same decision.

(23:48 - 24:11)

And over time, maybe we adjust that formula, right? And we improve it and we get a little bit more precise and nuanced as we see that some decisions are better than others. So we actually went ahead and built it because that's what we do here. And yeah, so if you wanna try this out, you can give us a shout.

(24:11 - 24:26)

I didn't mean for that to rhyme, but nevertheless it did. So give us a shout, Sean at Avergence.team. We'll try to keep the sing song to a minimum. And are you people is the question or are you an agent? I don't know.

(24:27 - 24:33)

We'll talk to you either way. Everyone's welcome here. So that's it.

(24:34 - 24:44)

Thanks for your time. I'll publish this and yeah, appreciate any comments, feedback. If you're working on similar stuff, love to talk shop and see where we go from there.

(24:44 - 24:45)

All right, have a great day.