Magic Tricks, Moats, and the Three-Body Problem of AI Networks

We’ve been building an AI-native network business at SuperMe for the last year and a half. One question that struck us near the beginning of this journey and more recently with OpenClaw, and that we’ve since been asked by other operators and investors: why isn’t anyone else? Behind that question are two deeper assumptions:

• AI should be transformative enough to create a new wave of consumer network products

• Those product ideas should be attractive to entrepreneurs

When you look at what’s happening in the market and compare it to previous platform shifts, the answer becomes clearer. New network businesses require patience, new land, and a very specific alignment of variables. None of which are enabled by the current environment.

Magic Tricks vs. Moats

During a platform shift, it’s common for founders to seize new capabilities to create magic tricks: experiences that feel novel because they lean on new tech, not new structure.

These products often:

• Go viral (or seem to)

• Use undifferentiated infrastructure (e.g., commodity LLMs)

• Have low retention

• Are easy for competitors to replicate

Viral growth alone isn’t a moat. Once the trick is explained, or once a big incumbent and startups clone the behavior, the product becomes a commodity. And commodities don’t compound.

This dynamic played out during the mobile platform shift too. App Store favorites featured in early iPhone commercials and viral hits drove spikes in downloads, but the companies that endured actually came a few years later. Instagram, TikTok, Uber, DoorDash, and Snapchat were not magic tricks. They were architected for defensibility from the start.

AI products today are repeating the pattern: impressive initial growth, shallow retention curves, and little unique ownership of a user’s attention or behavior. Unlike mobile startups of the past, many of these AI tricks monetize early, and must do so to pay for compute. This allows them to brag about ARR instead of downloads, but it only masks the underlying business’s weakness for slightly longer than viral apps during the App Store boom.

When entrepreneurs can do the cheap thing and reap all the rewards short term, why worry about the long term? It’s hard to understand, internalize or admit when you’re going so fast that you’re about to drive off a cliff. Growth momentum makes cliffs hard to see.

Even products intended as feeder experiences into bigger, more defensible plays often end up trapped in act one. Teams struggle to sustain growth in act one instead of investing in long-term defensibility of act two because act one is scaling so quickly.

Why Venture Capital Still Funds This Stuff

If the ARR isn’t eye popping enough for you with the first generation of AI companies, the valuations should be. Are VC’s just getting distracted by high flying growth? Shouldn’t they know better? You can’t blame investors for backing fast growth. Venture capital is a portfolio business, not a concentration business. For a VC, If they back ten fast-growth companies and one or two turn that growth into a sustainable business, they are one of their generation’s best investors.

Founders and employees only get one shot at a time. That misalignment of portfolio construction vs. concentration is part of why so many startups chase growth signals that look good in a pitch deck but don’t translate into durable user value as measured on a long term time horizon.

The last decade of venture has been dominated by SaaS. The muscle memory of most investors today is subscription metrics, not network density. That shapes what gets funded and how it’s evaluated.

Yes, HALOs and acquihires soften some downside risk. Yes, early secondary markets let founders and early employees capture liquidity. But those outcomes require specific talent and timing that most early-stage teams won’t have.

No New Land

Alex Zhu, the founder of Musically (which became Tiktok), gave the best explanation for the raw ingredients required for new networks to arise. Alex mentioned that building a new consumer network product is like building a new country. And it’s a lot easier to build a new country when there is new land that opens up. Mobile opened up a lot of new land in terms of internet access on the go. Instagram, Tiktok, and Snapchat consumed a bunch of free time people had waiting for coffee, at the DMV, etc. AI hasn’t yet opened up new land, which means as consumer founders we need to steal time and attention from some other product, one that is frequently extremely well optimized at keeping its users’ attention.

New land often creates new distribution channels. Consumer networks, which monetize slowly, depend on those cheap acquisition loops to reach density. Distribution tends to lag capability.

To be clear, not every consumer wave requires a new platform. In venture they call this the “why now.” Sometimes the unlock is regulatory, not technological. The deregulation of online gambling has created a surge of consumer network-oriented startups such as Kalshi, Polymarket, Underdog, Whatnot, Whop and others without any new land. The “why now” was a combination of legal permission, innovations in variable reward mechanics, and video infrastructure.

AI, so far, has increased capability. But it hasn’t yet created a comparable structural unlock for new consumer networks.

The Network Three-Body Problem

The final dimension preventing consumer network growth is structural: consumer network products don’t have a simple product/market fit geometry. They have a three-body problem.

Network product/market fit isn’t one variable. It’s an alignment problem.

1. The right target audience

2. The right network size

3. The right value proposition

If any one of these is off, you can’t tell which one is broken just by looking at the metrics. And if you misdiagnose early churn as “not enough users,” you’ll misallocate your budget to acquisition instead of fixing the real problem.

Marketplace and SaaS founders get clearer feedback loops:

• In SaaS, you tune value prop.

• In marketplaces, you tune liquidity.

• In networks, you’re tuning three variables at once.

That’s why some products burst into existence, spike in growth, and then fizzle, and why building a defensible network takes multiple cycles of learning, not just initial virality.

–
Until AI creates new land, or someone finds a way to manufacture density without it, most attempts will default to magic tricks to go viral and not create lasting value. The opportunity isn’t in the trick. It’s in compounding trust and density over time. The first wave of AI products optimized for speed. The second wave will optimize for structure.

Feel free to ask Casey AI more about this topic.

Currently listening to my Best of 2025 playlist on Spotify.