The State of AI in Kansas City in 2026: From Chat Wrappers to Decision Systems

Oct 8, 2026 · @Brian Curry

Kansas City’s logistics, manufacturing, agriculture and animal health companies are done paying for chat wrappers. What replaces them, from the founder of KC AI Lab and Vector1 Research.

By Brian James Curry, Founder of Vector1 Research, Kansas City

The experimental phase of generative AI is over. For two years the market filled up with lightweight API wrappers: chat interfaces sold as enterprise products. In 2026 the reckoning has arrived, and it is sharpest here in the Midwest.

Kansas City’s economy runs on freight, heavy manufacturing, agriculture, animal health and corporate retail. The executives running these tight-margin, asset-heavy balance sheets have learned that summarizing PDFs does not change a P&L. Demand has moved from generative novelty to quantitative decision execution: systems that choose, act, and can show the measured effect of what they did.

I have watched that shift from inside Kansas City’s AI community since 2016, when I started the group that became KC AI Lab. Over those years I saw startup after startup pitch software for problems that were really data architecture and econometrics problems. Before and alongside that, I spent more than twenty years leading analytics, data science and ML work at enterprises including Hallmark, Garmin, Koch Industries, Tractor Supply Company, Vail Resorts, AT&T and McClatchy.

That mismatch is why I founded Vector1 Research, an independent applied AI lab based in Kansas City. Large, traditional enterprises do not need another SaaS subscription. They need data science and AI strategy grounded in math, economics and how their operations actually run.

This piece is my read on where Kansas City stands in 2026, drawn from ten years inside its AI community: how the region moved from chat wrappers to decision systems, what those systems look like in practice, the open-source tools I have built for them here, and what the new hyperscale data centers mean for local companies.

Ten Years of AI in Kansas City, 2016 to 2026

Kansas City’s AI story runs in three phases: a practitioner community built around classical machine learning, a wave of generative wrappers, and now a turn toward decision systems. I have been part of all three.

In 2016 I started the Kansas City Machine Learning Group. The early work was the unglamorous core of the field: forecasting, classification, experimentation, and the hard problem of getting a model into production and keeping it there. The group drew data scientists and engineers from the region’s large employers, along with students and founders, and it grew into KC AI Lab, an AI and machine learning education community that passed 1,500 practitioners and became one of the largest community-built AI organizations in the Midwest.

By 2018 we had organized KC AI Lab as a member-owned education and development cooperative. Committed members trained for free, taught by people working in AI on real-world data, and the lab took on problems for businesses and individuals who had no AI resources of their own. That April, Startland News covered the lab and my argument that organizing Midwest talent around machine learning could be aimed at problems like income inequality, and that AI would force a rethink of “jobs, income, capitalism.” Eight years later that is still my research agenda, now in The Autonomy Economy and in essays like The Ownership Question.

Through those same years I was doing the work inside enterprises: leading analytics, data science and ML teams, building experimentation and causal measurement programs, and running cloud ML operations across Azure, AWS and Google Cloud at companies with revenues from $4 billion to $125 billion. That dual view, community on one side and enterprise P&L on the other, is what shaped how I read the region.

After ChatGPT launched in late 2022, the community changed almost overnight. Attendance jumped, and so did the pitches. From 2023 through 2025 the market filled with products that were a prompt and a user interface on top of someone else’s model. Some were useful. Most did not survive contact with a Kansas City operations team.

By 2025 the infrastructure itself started arriving: Meta’s Kansas City data center came online, and in February 2026 Google confirmed a campus of up to $10 billion in the Northland. And in 2026 the question local executives ask has changed. They no longer ask what a model can say. They ask what it can decide, and how they will know it decided well. Answering it is the work Vector1 Research was built for.

Why the Wrapper Era Ended Here First

Kansas City’s core industries are decision-dense, which is why optimization and causal measurement pay off here faster than chat tools do.

Start with freight. KC SmartPort, the logistics initiative of the Kansas City Area Development Council, describes the metro as the largest rail hub in the U.S. by tonnage, with four Class I rail lines, four intermodal parks, and two-day reach to 90% of the contiguous U.S. (KC SmartPort). Every one of those shipments carries routing, carrier selection, dock scheduling and inventory placement decisions. Small percentage gains on decisions repeated millions of times are where the money is. I wrote a practical companion to this on geospatial analytics and route optimization.

Then animal health and agriculture. The KC Animal Health Corridor stretches from Manhattan, Kansas, to Columbia and St. Joseph, Missouri, and its more than 300 companies account for 56% of worldwide animal health, diagnostics and pet food sales (KC ADC). Pricing, demand planning and channel measurement in those businesses are causal questions, not forecasting questions.

Retail and consumer brands round it out. Regional headquarters and distribution networks face the same pressure I described in The End of Scale: AI lowers the cost of coordination, which erodes the advantage that size alone used to buy.

This is personal ground for me. I grew up in a small Kansas farming town, worked crop fields from age twelve, and went on to Kansas State University. The operators I meet at regional events think in yields, margins and lead times. They want AI that speaks that language.

What a Decision System Looks Like

The enterprise value of AI in 2026 sits in multi-agent systems that execute work inside the systems a business already runs. We have moved past the chatbot era, where models answered questions, and the copilot era, where they suggested actions. This is the era of autonomous execution.

Think of a multi-agent workflow as a digital assembly line. A procurement agent does not just recommend a vendor from historical data. It reads from the warehouse in Snowflake or Databricks, vets the supplier against contract terms, and creates the purchase order in the ERP. A second agent reconciles the receipt. A third flags price drift. Value appears only when the agents are integrated with the systems of record. An agent that writes a recommendation into a chat window is a copilot with extra steps.

Human-in-the-loop versus human-on-the-loop

This requires a governance shift. In a human-in-the-loop workflow, an operator approves every step. In a human-on-the-loop design, people set the boundaries and economic parameters, and they handle the high-stakes exceptions, while agents execute the volume.

A worked example makes the difference concrete. Take a Kansas City distributor processing 4,000 purchase orders a month, with a buyer spending 12 minutes reviewing each one. That is 800 review hours a month. Now let an agent auto-execute any order that passes three guardrails: an approved vendor, a price within 3% of contract, and a total under $25,000. If 85% of orders pass, review time follows review hours = N × (1 − a) × t, where N is order volume, a is the auto-execution rate and t is review time per order. That gives 4,000 × 0.15 × 12 minutes, or 120 hours. The buyers recover 680 hours a month, and they spend the hours they keep on the orders that actually carry risk. These figures are illustrative; your own rates will differ, and the formula is the part to reuse.

The catch is data quality. An agent acting on a stale vendor master or an unmapped product hierarchy will execute the wrong decision at machine speed. Designing these guardrails, and the memory and audit trail behind them, is a core focus at Vector1. Agents need to remember what they did and why before anyone should let them do more, which is the problem behind my memory research described below.

Correlation Is Not Strategy: Why Causal Inference Is the Enterprise Mandate

Models that learn only from historical correlations break when markets move, and they cannot tell an executive which intervention to make. In agriculture and supply chain logistics, predicting a disruption or a demand drop is not enough. You need to know which action fixes it.

Causal inference answers counterfactual questions. If we change the pricing model, reroute a lane, or shift marketing spend, what will actually happen? Standard reporting misattributes value because the places where we act are rarely random.

A worked example

A regional retailer runs a promotion in its stronger stores. Promoted stores sell 1,240 units a week; unpromoted stores sell 1,000. The dashboard reports a 24% lift. But the promoted stores were already selling 1,150 units before the promotion, and the unpromoted stores rose from 1,000 to 1,020 over the same weeks because of seasonality.

A difference-in-differences estimate separates the two: DiD = (Y_T,post − Y_T,pre) − (Y_C,post − Y_C,pre), where T is the treated group and C is the comparison group. Here that is (1,240 − 1,150) − (1,020 − 1,000) = 70 units, about 6.1% of the promoted stores' baseline. The dashboard over-credited the promotion by roughly 3.9 times. Multiply that error across a marketing budget or a pricing program and it decides whether a strategy gets funded.

Difference-in-differences rests on an assumption, that both groups would have trended alike without the promotion, and that assumption should be tested rather than trusted. Good causal work does not promise certainty. It gives defensible estimates with honest uncertainty, which is far more than a correlation can offer.

Open Source Built in Kansas City

Much of what I have learned about decision systems is public, as open-source Python projects on GitHub and research papers on Zenodo. Each one started as a problem I kept seeing in enterprise work and could not find a good tool for.

Papilon is my most general framework: machine learning, causal modeling, Bayesian marketing mix modeling, simulation of economic systems, and decision optimization in one place. It exists because marketing and operations teams kept asking measurement questions that their attribution dashboards could not answer honestly.

PyCausalSim is a simulation-driven framework for causal discovery and evaluation. You simulate systems where the true causal structure is known, then check whether a method recovers it before trusting that method on your real data. It is the test harness I wanted every time someone showed me a causal graph with too much confidence.

MeaningFlow is a semantic intelligence framework that uses embeddings, clustering, dimensionality reduction and graph methods to model content and build knowledge graphs.

Daedalus maps money flows and runs causal analysis on economic systems. It sits alongside my research framework The Autonomy Economy, a macroeconomic model of how AI-driven automation affects firms, labor and capital.

tincast is an MIT-licensed forecasting framework for commodities that AI chip production depends on, starting with tin. It tracks a daily fragmentation index of regional price gaps created by export controls and supply-chain splits (write-up).

Memory-Node Encapsulation and the NeoCortex-M architecture are my research into artificial episodic memory and causal reasoning in agent systems: giving agents a structured record of what they did and why. It is the research counterpart to the human-on-the-loop governance described above.

All of it was built in Kansas City, and all of it is meant to be used. If you run one of these on a real problem, I want to hear what broke.

The Compute Is Arriving. The Decision Layer Is Not.

Kansas City has become a hyperscale data center market, and that changes what local enterprises can build, but not by itself.

In February 2026, Google confirmed Project Mica, a campus of up to $10 billion on roughly 500 acres in the Northland near I-435 and Highway 169, planned as five buildings and drawing up to 500 megawatts at full capacity (Axios Kansas City). Meta’s Kansas City data center, an investment of more than $1 billion, is already operational and serving traffic (Meta). Power is now part of the regional AI story too: Google’s campus falls under Evergy’s large load tariff, which requires long-term contracts for loads of that size.

Those buildings will power global cloud and AI services. They will not decide which lane a Kansas City carrier should drop, what an ag cooperative should charge next season, or whether a retailer’s promotion actually worked. Compute is becoming a commodity that arrives by fiber. The scarce part is the layer above it: clean operational data, decision models that encode how a business makes money, and people who can tell a sound estimate from a confident one.

That is where I expect the regional opportunity to sit for the next five years. Construction brings trades jobs and a few hundred permanent roles. The larger economic return comes when the region’s existing employers turn cheap, nearby compute into better decisions, and when the engineers trained in local communities build the systems that do it.

The appetite is there. A Heartland Forward poll found Kansans have the second-highest level of AI optimism in the heartland, while small businesses across the region still cite cost, skills gaps and trust as their main barriers (Startland News). Closing that gap is a data and talent problem, not a hardware one.

Kansas City’s AI Community Has Grown Up

Kansas City’s AI scene has moved from wrapper demos to production systems, and its community is a large part of why.

KC AI Lab’s 1,500-plus members have always been a cross-section of the region: data scientists from the region’s large employers, engineers from local startups, students, and operators who wanted to know what the models could do for their business. Today the city has several active groups, including the Kansas City AI Club and the Global AI Kansas City chapter, and the conversation in all of them has turned from what is possible to what is working.

That maturity will be on display at the Kansas City AI Expo on November 10 at The Mint in Lenexa. What makes the event worth attending is what it leaves out: no hollow stage program, just a floor of founders, operators and technical leaders showing systems that run in production. I will be there, and I am glad to talk agentic architecture or causal measurement with anyone who wants to compare notes.

Where Chief Data Officers Should Start

Audit your data architecture before you buy another AI product. You cannot build agentic workflows or causal models on fractured data foundations. Five questions will tell you where you stand:

1. Can you name the system of record for vendors, products, customers and prices, and do the agents you plan to deploy read from it?
2. For your top ten recurring decisions, do you know the volume, the cost of an error, and the share that could run inside guardrails?
3. When you credit a campaign, a price change or a new route with a result, was that result measured against a comparison group?
4. If an agent took a wrong action last Tuesday, could you reconstruct what it saw and why it acted?
5. Do you have people, in house or through a partner, who can test a model’s assumptions before it touches a live decision?

If more than two answers are no, the first project is the foundation, not the model.

Frequently Asked Questions

Who is Brian James Curry?

Brian James Curry is a Kansas City AI researcher and data scientist, the founder of Vector1 Research and of KC AI Lab. He has led analytics, data science and machine learning work for more than twenty years and has been building Kansas City’s AI community since 2016.

What is Vector1 Research?

Vector1 Research is an independent applied AI lab in Kansas City, founded by Brian James Curry. It works on agentic AI systems, causal inference, marketing economics and knowledge engineering for traditional enterprises.

What is KC AI Lab?

KC AI Lab is a Kansas City AI and machine learning education community. It began in 2016 as the Kansas City Machine Learning Group, was organized as a member-owned cooperative by 2018, and grew to more than 1,500 members.

How is AI in Kansas City changing in 2026?

Kansas City enterprises are moving from generative chat tools toward decision systems: agents that execute work inside core business systems within set guardrails, and causal models that measure the effect of each decision. Hyperscale investment, including Google’s Project Mica in the Northland and Meta’s Kansas City data center, is adding compute capacity to the region.

What open-source AI projects come from Vector1 Research?

Brian James Curry’s open-source projects include Papilon, PyCausalSim, MeaningFlow, Daedalus and tincast, published on GitHub under Bodhi8. They cover causal modeling, Bayesian marketing mix modeling, simulation-based causal discovery, semantic knowledge graphs, economic flow analysis and commodity forecasting.

What is human-on-the-loop AI?

It is a governance model where people set boundaries, economic parameters and exception rules, and AI agents execute routine decisions within them. It differs from human-in-the-loop, where a person approves every step.

What is the difference between causal inference and predictive analytics?

Predictive analytics estimates what will happen. Causal inference estimates what will happen if you take a specific action, which is the question most business decisions depend on.

About the Author

Brian James Curry is a Kansas City AI researcher and data scientist, the founder of Vector1 Research, an independent applied AI lab, and the founder of KC AI Lab, which began in 2016 as the Kansas City Machine Learning Group and grew to more than 1,500 members. He has more than twenty years of experience leading analytics, data science and machine learning at enterprises including Hallmark, Garmin, Koch Industries, Tractor Supply Company and Vail Resorts. His open-source work, including Papilon, PyCausalSim, MeaningFlow, Daedalus and tincast, is on GitHub, and his research is published on Zenodo. He grew up in a small Kansas farming town and attended Kansas State University. He writes about causal inference, agent systems and the economics of AI on Medium and Substack. Contact: brian@vector1.ai.