Most conversations about AI jump straight to frontier models and superintelligence. Russ Wilcox starts with a sewer. The CEO of ArtifexAI and advisor to the Pentagon and State Department joined Jeff Pedowitz on Unscripted to explain how mining boring public records led him from a Cape Cod zoning board to analyzing China's national AI strategy, and why he believes the same insight connects both: whoever treats AI as infrastructure, not magic, wins.
Wilcox is a self-taught programmer and computational physicist who never took a computer science class. An Army brat raised on the value of service, he was a full-time federal employee at the US Geological Survey at 16 and published research before he turned 18. He helped pioneer early language model work in Oslo, training small models on YouTube transcripts before most of the industry knew what an LLM was. Today he runs ArtifexAI, chairs the Policy Committee of the American Society for AI, and writes analysis on Chinese technology strategy read by defense and Fortune 500 leadership.
During COVID, Wilcox moved home to Cape Cod and found he couldn't swim in the ponds of his childhood. Algal blooms had ruined them. The culprit traced back to zoning, so he went to his first zoning board meeting and watched his town debate a $250 million sewer project on the advice of a consultant. Nobody could say where the data came from. Nobody had read the permits. The records existed, on paper, in boxes, unread.
So he built the tool himself: AI that digs through permits, meeting minutes, and decades of municipal archives, then maps where the town should actually invest. The town shrugged and listened to the consultant anyway. But Wilcox had found his company.
His core insight applies to every marketer and executive drowning in dashboards: structured data represents maybe 15 percent of human knowledge. The rest lives in documents, transcripts, videos, and meeting minutes that nobody reads. Humans like to write. They don't like to read. AI finally makes the reading scalable.
This is where the sewer meets the superpower contest. Wilcox argues the West and China are running two different races.
The West is building "god models": massive frontier systems you can ask anything, trained mostly on English-language data, disconnected from local context. China is treating AI as infrastructure. Provinces feed "data reservoirs" that power specialized swarm agents for planning, energy, and logistics. Five-year plans get built on unified data. Research initiatives in developing countries sit along 70 percent of the world's trading routes, and the data flows home.
His chip war analysis follows the same logic. While Washington slapped export controls on high-end chips, Wilcox mined public construction and corporate records and found the real story in chip packaging, the unglamorous technology that lets chips talk to each other. China controls roughly 90 percent of the Asian market, and he mapped three new packaging facilities under construction while the US was still funding research. Export controls target the wafer. China is building the system.
The race, in his framing, isn't about who builds the biggest model. It's about who raises the level of humanity with AI. Treating AI as the next industrial revolution, a pure efficiency play, is how you lose. If that argument sounds familiar, it should. It's the same one we make about marketing organizations every week.
Wilcox rejects the idea that the federal government will lead this. His bet is local, and his playbook has three parts.
Compliance. Local governments run 15 to 20 years behind on technology, and property taxes are their lifeblood. AI that unifies land records, flags unregistered short-term rentals, and streamlines permitting pays for itself fast.
Decision support. ArtifexAI built a tool it calls the Oracle, SimCity for real life. Simulate a school placement, an energy project, or a sewer line against decades of historical records and see the downstream effects before spending a dollar. One state client had ArtifexAI mine five million pages of other states' regulatory hearings to learn what succeeded and what failed before launching its own infrastructure project.
Visibility. Town managers have no KPIs because the data is trapped in unstructured records. Unlock it, and civic leaders finally see what's working.
The most transferable part of the conversation was about adoption, and it will sound familiar to anyone selling transformation. Wilcox's rules:
Don't lead with "AI." His team describes decision support tools built on the community's own data, transparent and traceable. Vocabulary kills projects before they start.
Start small with a real problem. Solve the short-term rental crisis or the permit backlog first. Earn the right to expand.
Be transparent about data. Your data is an asset. It stays yours. It doesn't get mined, resold, or fed into someone else's model. Trust is the product.
Acknowledge dual use. The same technology that finds every unpermitted septic system can find every time a citizen complained on public record. Naming the risk, and governing it, is what separates partners from vendors.
Wilcox sees a fork. One path is what he calls digital techno-feudalism: a handful of frontier model companies mediating human knowledge. The other is sovereign, community-level AI that makes local government transparent, helps neighboring towns learn from each other instead of fighting, and lets communities prepare proactively for the energy and water crunch he sees coming. He believes local democracy isn't dead, just stagnant, and that public-private partnership, not government alone, revives it.
What does ArtifexAI actually do?
It uses AI to convert unstructured public records, permits, meeting minutes, and hearing transcripts, into structured data that local governments and companies can use for compliance, planning simulations, and risk modeling.
Is Wilcox anti-China?
No, and he's explicit about it. He calls himself an analyst, not an apologist: he believes democracy is the best form of government and that the US can win, but only by learning from how China treats AI as a system rather than a product.
What's the difference between "god models" and swarm agents?
God models are giant general-purpose LLMs trained to answer anything, with hallucination risk and limited local context. Swarm agents are smaller, task-specific models trained on curated structured data, the approach Wilcox argues delivers more reliable real-world outcomes.
What should business leaders take from this episode?
The same lesson we preach in revenue marketing: unstructured data is your biggest untapped asset, efficiency alone is a losing strategy, and adoption is a change management problem before it's a technology problem.