Network-Centric Investing: When Second Brains Learn to Coordinate

Seeing the present more clearly than everyone else

In 1997, I sat in a classified facility with 100 people asking a question that would reshape American military doctrine: How do we coordinate U.S. power across domains so that a special ops team can put a rose on Saddam Hussein’s pillow, and Saddam wakes up knowing they could have put a bullet there instead?

That question became Network-Centric Warfare. The answer took decades to implement. The capability you saw a few weeks ago in Venezuela, where we extracted a dictator, traces back to that room. A few months of planning and one intense night operation.

The same architecture now applies to capital markets. And a viral open-source project called OpenClaw accidentally revealed why.

The Coordination Problem

The military problem in 1997 was simple to state and brutally hard to solve.

The Navy doesn’t talk to the Marines. The Federal Reserve doesn’t coordinate with special operations. Media doesn’t talk to either. Intelligence agencies hoard signal. Each domain optimizes locally.

But certain missions require all of them, in sequence, timed precisely. You need the Treasury to freeze accounts before the operation. You need media to shape the information environment after. You need the power grid to go dark at a specific moment. You need extraction routes cleared by assets no one knows exist.

No single domain can execute this alone. The problem isn’t capability. It’s coordination.

In 1990, invading Baghdad to extract Saddam would have required a million troops. By 2002, we did it with 120,000. The difference wasn’t better tanks. It was better information architecture. We learned how to get different domains to see each other’s signal and synchronize action.

Hayek identified the same problem in economics 80 years ago.

The knowledge needed to allocate resources optimally exists. It’s dispersed across thousands of people who don’t know what each other knows. The farmer knows local soil conditions. The grocer knows which products move. The clinician knows which interventions work. The founder knows which technology is maturing.

No central planner can aggregate this knowledge. By the time you collect it, it’s stale. The signal is distributed by nature.

Hayek’s insight was that markets solve this through price. Price compresses distributed knowledge into a single number that coordinates action without anyone needing to understand the whole system. They are a form of information.

But prices are a lossy compression. They tell you what happened. They don’t tell you what’s emerging.

The Information Arbitrage

In the mid-90s, when banks fled Silicon Valley after the savings and loan crisis, Hambrecht & Quist doubled down. They weren’t smarter than Goldman. They were better positioned.

H&Q sat in the middle of the startup stew. They saw deal flow before anyone else. They knew which engineers were leaving which companies. They knew which technologies were actually working versus which were just funded. They had lunch with the founders building the future.

This wasn’t analysis. It was information arbitrage. Their presence in the network exposed signal that outsiders couldn’t access.

Dan Case, who ran H&Q, understood something most investors still don’t: the value isn’t in predicting the future. It’s in seeing the present more clearly than everyone else.

When you’ve looked at 6,000 startups in a sector, talked to 10,000 people trying to solve related problems, you don’t need to predict. You can see. You know which problems are actually getting solved. You know which solutions are three years away versus ten. You know where the real friction lives.

This is the knowledge Hayek described. Dispersed. Local. Perishable. Impossible to centralize. But extraordinarily valuable to whoever can access it.

Second Brains

Now something new is possible.

OpenClaw is an open-source AI agent that went viral last week. It runs locally on your machine, connects to your messaging apps, and acts autonomously on your behalf. It reads your emails. It manages your calendar. It remembers your conversations. It executes tasks while you sleep.

It might take us a week to process a DocuSign that requires a few obscure account numbers to complete. That is child’s play for OpenClaw.

In one case, a user asked OpenClaw to make a dinner reservation. Unbeknownst to the user, the restaurant was not on OpenTable. So OpenClaw, without asking, went to GitHub, found a code base for making calls, wrote an app so it could speak, called the restaurant, and booked the reservation by voice.

Skynet?

The security implications are terrifying, which is why I wouldn’t run it on anything touching fund operations. But the capability it represents matters more than the specific tool.

I have a copy running Ubuntu/WSL on my Windows box, in a Docker instance. Sort of like being locked in a BSL-4 facility protecting a gain-of-function research project. I figure for right now, that might prevent it from getting loose. I can chat via WhatsApp. I don’t have my email set up yet, but I'll probably do it soon. I gave it some tasks to monitor social media. I did ask it to find a used ThinkStation capable of running OpenClaw on its own Linux box. It shopped for its new home and found a couple of used units for about $400, good enough to do the job (i7 Gen 10, 16 GB, 256 GB NVMe. Luckily, it did not have my credit card.

For the first time, individuals can externalize their knowledge into a persistent, queryable, actionable system. A second brain that doesn’t just store information but can act on it.

I’ve been building something like this for our investment work. Upload your healthcare records, your Oura data, your grocery receipts, your research notes. The system assembles context across sources. It sees patterns you’d miss. It remembers conversations from six months ago that suddenly matter.

But we are building much more.

What happens when second brains can talk to each other?

The Network Layer

Imagine 100 people working on reversing chronic disease. Researchers tracking metabolic pathways. Clinicians running interventions. Founders building measurement tools. Investors evaluating the space. Policy people navigating regulatory barriers.

Each one runs a second brain. Each agent knows what its principal knows. Their contacts. Their insights. Their open questions. The papers they’ve read. The conversations they’ve had. The problems they’re stuck on.

Now connect them.

Not a Slack channel where humans exchange messages. A network where agents query agents on behalf of their principals.

• “Who in this network has seen promising data on GLP-1 effects in NAFLD?”

• “Anyone tracking regulatory movement on ultra-processed food labels?”

• “Who has relationships at FDA’s nutrition division?”

• “What companies are solving the cold chain problem for regenerative produce?”

The knowledge exists. Right now, it’s trapped in individual heads and inboxes. Ellen and Sherry are transforming Food/Health. Ellen met Sherry at Katie’s conference. Sherry met Katie because Sherry and I happened to sit next to each other on a bus at the Farmers/Ranchers conference. I thought Sherry should work with Katie. These serendipitous collisions are how networks actually form. But the collision rate is pathetically low given the knowledge available.

Agents could route around that.

Hayek Meets Shannon

Claude Shannon, the father of information theory, gave us the mathematics of signal and noise. A communication system has a source, a transmitter, a channel, a receiver, and a destination. Noise corrupts the signal in transit. The question is how much information can reliably flow through a noisy channel.

Hayek’s coordination problem is Shannon’s communication problem applied to economics. The signal is distributed knowledge about local conditions. The noise is everything that prevents the signal from reaching the people who need it. The channel capacity is limited by the friction of human communication.

What happens when you upgrade the channel?

When agents can query agents, you’re not just reducing friction. You’re changing the architecture. Each node transmits local signal. The network filters noise. Patterns emerge that no individual node could see.

This is what we built for Network-Centric Warfare. Different capabilities, different domains, different chains of command. But a shared information architecture that let them synchronize against objectives that no single domain could achieve.

The military version took billions of dollars and a decade to build. The civilian version might emerge from open-source tools and a protocol layer that doesn’t exist yet.

Frankly, I think NSA built OpenClaw to just stop with the pretense. Who cares if we walk around naked if it ends chronic disease?

What OpenClaw Actually Reveals

OpenClaw previously branded itself around the concept of “exfoliate.” In cybersecurity, exfiltration is the unauthorized transfer of data from a system. Exfoliation is the stealthy version: peeling data in thin, incremental layers over time to avoid detection.

This is not a bug in their architecture. It’s the value proposition.

And that tells you something important about where this is heading.

The ability to slowly extract signals across system boundaries, synthesize them, and act on them is exactly what agent-to-agent coordination requires. Your agent needs to query my agent. My agent needs to return something useful without dumping my entire knowledge base. The exchange has to be bounded, permissioned, and incremental.

OpenClaw solved the agent problem by making exfoliation a feature. They did not solve the trust problem that makes exfoliation safe.

This is the gap.

The same capability that lets your second brain coordinate with mine is the capability that lets a malicious actor peel your knowledge base layer by layer without triggering alarms. The architecture that enables Network-Centric Investing is the architecture that enables network-centric extraction.

Cisco’s security researchers already found that 26% of OpenClaw skills contained at least one vulnerability. Malware authors have adapted their tools to specifically search for OpenClaw configuration files. The attack surface isn’t theoretical.

So the question isn’t whether agents will coordinate. They will. The economics are too compelling.

The question is whether the coordination layer gets built by people who understand that exfoliation cuts both ways, or whether it emerges from tools optimized for capability without constraint.

Network-Centric Investing

Here’s the investment thesis buried in all this theory.

We’re putting together a $100 million strategic venture fund focused on Food is Health. The investors are the supply chain itself: ingredient companies, CPGs, grocery, hospital systems, payers. Each brings capital, but more importantly, each brings signal. Costco sees purchasing patterns. Novo Nordisk sees clinical outcomes. Ingredion sees what reformulations are technically feasible.

The fund invests in technology that reveals human-based outcomes. Tools that show consumers actually want nutritionally dense food, which unlocks CPG investment, which pulls demand through the supply chain, which makes regenerative agriculture economically viable.

But the capital is almost secondary to the network.

When the supply chain shifts from horizontal to vertical integration, the rate of change is constrained by how quickly relationships can form between nodes that have never interacted before. If farmers grow more nutritionally dense crops, they need grocers who care, which means distributors who can handle it, which means consumers who will pay for it.

The diffusion rate is a function of how quickly personal networks rewire from the old architecture to the new one. (See molting or exfoliate)

Crusonia, the network we’re building alongside the fund, exists to accelerate that rewiring. Daily content builds the audience. Events create collision opportunities. The 6,000 deals we’ve evaluated become a knowledge base that lets us tell any node in the network: here’s what’s possible, here’s the maturity level, here’s where the friction lives.

Now imagine that knowledge base isn’t just accessible to our team. Imagine every serious player in Food is Health has a second brain. Imagine those second brains can query each other, with appropriate permissions, to surface signal none of them could see alone.

Note to Geeks: Docker instances with a forced YAML config wrapped in a private tailscale vpn network between network partners. Not perfect, but good enough.

A hundred early adopters, synthetically fused into a coordination layer that reveals market optionality before incumbents see it.

That’s Network-Centric Investing.

What’s Missing

OpenClaw gives you the agent. It doesn’t give you the protocol for agents to coordinate without their principals getting burned.

Three problems remain unsolved:

The coordination layer. When your agent queries my agent, what’s the grammar of that exchange? How do agents negotiate scope? How do they synthesize partial information from multiple sources into actionable insight? This is protocol design, not tool selection. Someone needs to build the TCP/IP of agent-to-agent communication.

Permission architecture. What am I willing to let others’ agents discover about my knowledge? There’s a difference between “I know something relevant” and revealing what I know. The system needs consent models, trust hierarchies, and boundaries that let agents signal without leaking. This is the hardest problem because it’s not just technical. It’s institutional. Incentive structures have to make contribution rational even when others might free-ride.

Selection bias correction. Connected networks are susceptible to groupthink. If everyone’s second brain is trained on similar inputs and optimizes for similar objectives, you get convergence, not intelligence. The system needs mechanisms to surface dissent, weight contrarian signal, and avoid the failure mode where the network becomes an echo chamber with better bandwidth. Diversity of perspective has to be architecturally preserved, not just hoped for.

The Protocol Is the Product

In 1997, we didn’t build a weapon. We built an architecture that let existing weapons coordinate. The F-117 and the SEAL team and the satellite and the cyber capability all existed. What didn’t exist was the information layer that let them synchronize.

The same is true here. The agents exist. OpenClaw, Claude, the various second-brain tools proliferating across the market. What doesn’t exist is the protocol that lets them coordinate without destroying the trust and permission structures that make the network valuable in the first place.

Whoever builds that protocol captures the coordination layer for Network-Centric Investing. Not a fund. Not a tool. The infrastructure that lets distributed knowledge synchronize against shared objectives.

The knowledge to solve chronic disease, to rewire the food system, to allocate capital to its highest use, already exists. It’s distributed across thousands of people who don’t know what each other knows.

The agents can see it now. They just can’t talk to each other yet.

Not yet.