The Fire Chief Doesn’t Need a Better Map. She Needs a Second Brain.

Geospatial data gave us the most powerful sensing layer ever built. Here’s why that’s only the beginning.

It’s Tuesday morning and a wildfire is moving toward three communities in Northern California.

The fire chief has more information than any fire chief in history. Satellite feeds updating every 30 seconds. Wind models. Fuel load data. Infrastructure maps. Evacuation route overlays. A real-time sensor mesh that her predecessors couldn’t have imagined.

She is drowning in it.

What she needs — what she has never had — is something that can reason through it with her. Not process it. Not display it. Reason about what happens next, under different interventions, before she commits her resources to a course of action she cannot reverse.

That is the Causal AI problem. And geospatial is where it is going to be solved first.

What Geospatial Got Right and What It Has Left Unfinished

Decades of geospatial investment has produced many remarkable advances: Sub-meter resolution. Near real-time planetary coverage. Models that can detect a building, classify a crop, identify a vehicle, estimate flood depth, from orbit, at scale, continuously.

This is the sensing layer. Incredibly powerful.

But this is a perceptual system, not a reasoning system. It can see the flood. It can tell you where the water is, how fast it is rising, which roads are already cut off. What it cannot do is tell you what caused the flood pattern you are seeing, what happens to it if the next storm cell arrives two hours earlier than forecast, or which is the best option to protect the most people given the available resources.

The gap between those two things, between seeing and reasoning, is not a data problem. It is not a compute problem. It is architectural.

Causal AI is what closes that gap.

The Three Layers

Think of the architecture in three layers, each built on the one below it.

The sensing layer is what the geospatial industry has built – what some are beginning to call planetary intelligence. Satellites, IoT networks, weather sensors, camera feeds; a continuous, high-fidelity picture of the physical world. This is the foundation. Without it, everything above is reasoning about abstractions.

The causal world model layer sits on top and transforms perception into understanding. Not just what is happening, but why, and what happens next under different conditions. A wildfire causal engine encodes the relationships between fuel load, wind speed, humidity, and topography that govern how fire actually behaves. It can simulate what happens next. It can test what would have happened differently. It can run the wind-shift scenario before the wind shifts.

The human decision layer is where this becomes consequential. And it is where almost every current AI deployment gets the design wrong.

Back to the Fire Chief

With the sensing layer alone, the fire chief receives information and makes decisions. The machine has accelerated what she can know. The causal heavy lifting or reasoning from that picture to the intervention most likely to save lives, is still entirely hers.

With a causal world model beneath her, something different happens. She doesn’t receive a map.

She receives simulated decision paths.

The causal engine has already run the wind-shift scenario. It has modelled what happens to the fire corridor if resources deploy here rather than there. It has ranked evacuation sequences by predicted outcomes; the lives protected, time required, resource cost. It has surfaced the three intervention options most likely to succeed given current conditions, with the reasoning transparent and reviewable.

She is not processing information. She is validating decisions the machine has already reasoned through, and making the final call with a quality of understanding no previous system could give her.

Her role has shifted from information processor to strategic authority.

The machine did the causal work. She brings what the machine cannot encode: judgment, accountability, the weight of consequences, the calls that are not causal questions at all … they are human ones.

Why This Isn’t Just a Geospatial Story

The fire chief’s problem might be geospatial, but the architecture it demands – sensing, causal reasoning, human strategic authority – applies everywhere decisions are made under uncertainty with high consequences and incomplete information.

Insurance: underwriters reasoning about risk that is physically manifesting in real time. Agriculture: growers making irrigation and harvest decisions against a causal model of soil, weather, and crop physiology. Supply chain: logistics planners simulating disruption scenarios before they cascade. Defense: analysts reasoning about intent, not just position.

In every case, geospatial data is the richest, most physically grounded source of real-world signal available. It is why the causal reasoning breakthrough is going to be built on a geospatial foundation, not because the problem is geographic, but because the data is grounded in physical reality.

LLMs were trained on what humans wrote about the world. Causal AI trained on geospatial foundations will be trained on what the world actually did.

That distinction will matter more than most people currently appreciate.

The Design Problem Nobody Is Talking About

Building the sensing layer is an engineering challenge. Building the causal world model is a scientific and engineering challenge of considerable depth. But designing the human layer, the interface between causal machine reasoning and human strategic authority, is something different.

It is a design problem, an organisational problem, and a trust problem simultaneously.

The fire chief needs to trust the scenarios the causal engine surfaces. That means the reasoning needs to be transparent and reviewable; not a black box producing recommendations she cannot interrogate. It means uncertainty needs to be visible. The system needs to know what it doesn’t know, and surface that honestly rather than filling gaps with false precision.

The people operating at this level need different capabilities, not deeper technical knowledge, but the judgment to govern systems they cannot fully see inside. Able to apply judgment at the level of strategy rather than execution. Able to ask the right questions of a machine that answers faster than any human team.

When the machine becomes more capable, this elevation does not happen automatically. It requires deliberate architecture. Deliberate organisational design. And a clarity about what the human is actually being asked to do that most current AI deployments have not yet achieved.

Closing the Loop

Geospatial data is giving AI its eyes. The richest, most physically grounded, most continuously updated picture of reality that has ever existed.

Causal AI is what gives those eyes a mind behind them.

The fire chief doesn’t need a better map. She never did. She needs a system that has already reasoned through her options before she asks, and can explain why.

That system is being built. The geospatial industry built its foundation without knowing it. The question now is who recognises that and moves first.

The broader implication is this: geospatial is not just one domain where causal AI will be applied. It is the domain where causal AI will be proven, because physical reality is the only ground truth rigorous enough to validate causal claims at scale. Insurance, agriculture, supply chain, defense – every vertical listed in this piece ultimately depends on what the physical world is doing. The sensing layer that reads that world is already built. The causal reasoning layer that understands it is what comes next.

That is what I call Causal Planetary Intelligence. Not a geospatial product. A reasoning architecture – grounded in physical reality, operating at planetary scale, with the human where the human belongs: in strategic authority, not information processing.

This is the third piece in a series on Causal Planetary Intelligence. The first argued that AI can perceive but not yet reason causally. The second traced the architectural gap between sensing and understanding. This piece opens the argument to Causal AI’s broader significance — with geospatial as the proof domain.

Matt Sheehan

Matt Sheehan is a geographer, AI strategist and senior executive with over 25 years of experience leading complex organizations through technology-driven transformation. His focus is decision velocity — compressing the distance between insight and action in environments where the cost of being slow is real. He is currently researching world models and causal reasoning engines as the next frontier of geospatial intelligence, and how these architectures can fundamentally augment human decision-making at scale. Reach him at mattsheehan@spatialnext.io

June 4, 2026June 4, 2026

Why AI Can See the Flood But Cannot Tell You What To Do About It

The causal reasoning gap — and why closing it is the foundation of Causal Planetary Intelligence

Last Friday I published an article arguing that Will Marshall‘s Planetary Intelligence framework, both compelling and consequential, is missing one critical architectural layer. Will and team are building the sensing system: the eyes, but what is not in place yet is the reasoning architecture that turns what those eyes see into genuine understanding.

I call this Causal Planetary Intelligence.

A few people responded to my post asking the same question but in different ways: what do you actually mean by causal? Why does the distinction matter? Can’t powerful enough AI simply figure out causation given enough data?

All great questions. Let me try to answer them in this follow-up article.

The two kinds of knowing

There is a significant difference between knowing that two things tend to happen together and knowing why one causes the other. An academic argument. Actually not. This is the difference between being able to describe something and reason about what to do next.

Let me colour that picture.

Umbrella sales and cold and flu rates are strongly correlated. Both peak in autumn. A sophisticated pattern-matching system, trained on enough historical data, would be able to identify this relationship. It can predict that when umbrella sales rise, flu risk rises with them.

So it sees the relationship, but would never suggest banning umbrellas for public health reasons. Why? Because it cannot reason about causation.

Now think about a child. Autumn is here, she feels colder weather is here. It is raining on her way to school. She watches her classmates sneezing. But this child has built her own world model. She knows that cold wet weather brings with it both umbrellas and winter illness. And that carrying an umbrella doesn’t give you the flu, but regular hand washing just might.

This child does not follow a process of pattern recognition, she uses causal reasoning.

The librarian and the child

Will Marshall uses a powerful image in his Planetary Intelligence essay. He describes current AI as like a brilliant scholar locked in a library, someone who has read everything ever written, can synthesise and reason across vast domains, but has never stepped outside into the real world. Somebody who has never felt gravity or watched a fire spread or seen water rise.

Spot on, but I wanted to push that image one step further. The Large Earth Models (LEM) Will describes solve the scholars problem of blindness to the world. LEM’s give AI the equivalent of senses. In other words grounding AI in real physical data rather than text alone. That is a significant advance.

But letting the scholar outside solves one part of the problem. But there is a second challenge. The scholar has read every book ever written about causation. They can tell you about flood dynamics, fire behaviour, the mechanisms of disease transmission.

But there is a difference between being able to describe causal reasoning and being able to do it.

The child who has dropped a ball. He felt the release, watched its path, heard the impact, but knows nothing about gravity. Unlike the scholar, he has built a working model of how the physical world works. He can reason and predict: what would happen if he threw it harder, or from a cliff, or into water.

LLMs are extraordinary scholars. They are not children. And the distinction matters enormously the moment you move from describing a problem to deciding what to do about it.

Where geospatial went wrong and where AI is compounding it

I’m a geographer; a discipline centred on causal explanation. Geographers are focused on the why: why do Big and Little Cottonwood Canyons here in Utah look so different, why did the Mormons settle in the Salt lake valley and not elsewhere, why does weather behave differently on this side of the mountain.

Pure causation.

Geospatial inherited geography’s territory but not its ambition. Where geography asked why, geospatial asked where. It became extraordinarily good at showing what is happening and where, and largely stopped asking why. Geospatial is a correlation business: Data collected, processed, rendered, handed to an analyst, then on to a decision-maker who looks at this actionable insight and makes the causal leaps themselves. The technology never had to understand. The human did the understanding.

Correlation is a statistical relationship between two variables that move together — when one changes, the other tends to change

Before AI, the geospatial division of labour was reasonable since humans were the only reasoners. But that was then, this is now. And here is the uncomfortable truth about AI in its current form: it has not solved this problem. It has made it more sophisticated.

At their core today’s AI systems are extraordinarily capable correlation engines. They can identify relationships across datasets at a scale no human analyst could approach. They can surface patterns that would take teams of researchers years to find. In short they are remarkable.

But they do not understand why those patterns exist.

They cannot reason about what happens when you intervene. They can tell you that X and Y tend to co-occur. They cannot tell you whether changing X will affect Y, or whether both are downstream of Z.

Large Earth Models extend this capability significantly. They are trained on physical sensor data rather than text. That means they can perceive the world in a way LLMs cannot. For example, they can tell you what a flood looks like today in your specific town, not just what floods look like in general. That is a big advance.

But perception is not understanding.

A LEM that has watched ten thousand floods can describe the next one with extraordinary precision. It cannot tell you what caused it, or what would have happened if the upstream reservoir had been managed differently, or what the optimal intervention is right now.

That requires a different architecture entirely.

The gap

There is much conversation about AI which is centred on data and processing power. Almost a fixation. That is all good, but correlation-based systems as we have today will never get us to causal understanding. We have an architectural constraint.

You cannot learn causation from correlation alone. You can learn that A tends to precede B. You cannot learn, from that observation alone, whether A causes B, whether B causes A, whether both are caused by C, or whether the relationship is coincidental. Distinguishing between these possibilities requires either experimental intervention; changing A and observing what happens to B, or a prior model of the mechanisms involved.

This is why world models are the necessary next step. Not an incremental improvement on current AI. A different architecture. One that encodes not just the state of the world but the mechanisms that connect states; the causal structures that allow reasoning about interventions and their consequences

This is what Yann LeCun and others on the cutting edge of AI have been arguing for. Not better pattern matching. A fundamentally different kind of machine reasoning.

Why this matters for anyone who makes decisions

Let me step back for a moment. The implications here reach well beyond AI research or the geospatial industry.

Every decision we humans make – in every board room, every clinical setting, every policy discussion, every operational headquarters – involves causal reasoning.

When you change a pricing strategy, you are betting lower prices drive volume. When you enter a new market, you are betting timing beats competition. When you respond to a threat, you are betting your move changes their next move.

At best, what AI currently offers decision-makers is a powerful description of the present and an extrapolation based on the past. In other words pattern recognition at scale. It cannot tell you what will happen if you intervene. Only what has tended to happen in the past.

Today’s correlation-based AI cannot reason about what will happen if you do this specific thing in this specific context.

Where this conversation going next

In my Friday article I introduced the architecture of Causal Planetary Intelligence, and the three layers that together close the loop Will Marshall’s framework points toward but doesn’t yet close. The sensing layer Will is building. The world model layer that reasons causally over what the sensors perceive. And the human layer — which is where the conversation about AI most frequently goes wrong.

This Thursday I will to go deeper into that architecture. Particularly the human layer, because I believe the framing most people bring to it – replacement versus assistance – misses what is actually most interesting and most important about what Horizon Two, the AI-powered paradigm shift, looks like for human decision-makers.

The question is not whether humans stay in the loop. They do. The question is whether AI becomes a genuine second brain — one that has already reasoned through the mechanisms, modelled the interventions, and arrives at the table with something more than a description of the past.

That is where the conversation will go next.

Matt Sheehan

Matt Sheehan is a geographer and AI strategist focused on one question: why does AI get better at seeing the world without getting better at understanding it? He is currently researching world models and causal reasoning engines as the next frontier of geospatial intelligence. Reach him at mattsheehan@spatialnext.io

June 2, 2026June 2, 2026

Planetary Intelligence Can See. It Cannot Yet Understand.

TL;DR: Current geospatial AI (including satellites, sensors, and Earth models) is great at telling us what is happening in real-time, but still relies on humans to figure out why and what to do. The author argues the missing piece is causal reasoning — AI that doesn’t just perceive the world but understands cause and effect, can simulate intervention scenarios, and hands humans pre-reasoned decision paths rather than just better maps. They call this “Causal Planetary Intelligence.

Will Marshall‘s recent Substack essay on Planetary Intelligence (PI) is the clearest articulation I’ve encountered of where geospatial is genuinely heading. It’s a must read.

I want to extend that conversation, because I think there’s one critical piece still missing from the architecture.

The pieces of the puzzle

Will describes Planetary Intelligence (PI) as a real-time sensing and computing system that continuously observes the entire physical Earth and converts that observation into actionable decision support for humans. He breaks PI down into 4 components:

Sensing: Continuous data streams – satellites, IoT, weather networks, cameras etc.
Modelling: Large Earth Models (LEM’s) trained on that sensor data, analogous to how LLMs were trained on internet text.
Edge Computing: Real-time data processing in space to reducing latency.
Decision support: The output that flows back to human decision-makers inside their OODA loop – Observe, Orient, Decide, Act.

As Will also discusses in The Spillover Podcast, this gives humans more and faster information for decision making. In other words a faster delivered picture of what is happening right now – situational awareness – and the insight to help those who need to act, take decisive action ‘before it is too late.’ This makes ‘the collective smarter than its parts’.

Will’s argument is centred on bandwidth to reduce decision latency. PI and LEM’s are critical foundations, but as I will argue, are the first stage. The critical bottleneck still in place is the decision making bottleneck, that is the second stage and what I will address here.

What Geography Always Promised

I’m a passionate geographer. This is a discipline centred on causal explanation, helping us understand better the physical world. We explore questions like: why is Big Cottonwood Canyon – here in Utah – V shaped while Little Cottonwood Canyon, 2 miles further south, U shaped? We observe, gather evidence, build and test our hypothesis. Geography’s final output has always been understanding, not action.

Geospatial – GIS and remote sensing – has long been associated with geography. Software and toolkits (platforms) which allow us to process, store, analyse and visualize location based data. The commercialization of GIS shifted that final output, actionable insight delivered to decision-makers by analysts became the end product.

AI in the form of machine learning (ML) and generative AI (LLM’s) have added new tools to the toolbox. But the current geospatial industry business model is centred on an analyst in the middle. What I call the current paradigm or Horizon One. Will indirectly touches on this in his framing, suggesting the human moves from processor to strategic authority. The question he leaves open is: what does the machine need to become for that elevation of the human to actually happen?

This is the real paradigm shift or Horizon Two.

From Correlation to Causal Understanding

When we were children, we learned by exploring and observing. That means using our senses to understand the world around us and learning from that interaction – letting go causes a ball to fall to the ground.

We were building our own world model.

As Will points out LLM’s are pattern matching against text, predicting relationships between words rather than understanding the world those words describe. They are the librarian who has read every book in the world, can expertly draw from what he/she has learned, but has never set foot in the physical world. Unlike a child, LLM’s have no understanding of causality, they can describe a ball falling in exquisite detail but cannot reason about why, or what would happen differently on the moon.

Geospatial has been predominantly focused on correlational not causation. There are exceptions including physics-based simulations, process models, but these have remained narrow, domain-specific, and disconnected from the broader world of decision-making.

LLMs compound this problem rather than solve it. They are the most sophisticated correlational engines ever built, pattern matching across the entire written record of human civilization, but they have no causal structure underneath. As Will discusses, they can describe a flood in extraordinary detail but cannot reason about what caused it or what happens next if you intervene.

LEM’s extend LLM’s. They are trained on physical sensor data, making them able to perceive the world, but crucially they do not reason causally.

Enter world models.

World Models

The next advance in AI will be the release of world models say Yann LeCun – sometimes described as the godfather of AI – and many others. A world model doesn’t just represent the state of the world, it represents how the world works. It encodes not just what happens, but why — and what would happen differently if you intervened

LEMs provide the perceptual foundation world models need. In other words the eyes. The world model is the reasoning architecture that transforms what those eyes see into genuine causal understanding. Without Will’s planetary sensing as input, a world model reasons about abstractions. Without the world model on top, LEMs perceive without understanding.

Article content — Geospatial AI Maturity Model

This is the architectural leap my diagram above captures. The LEM sits at The Leap – the boundary between Horizon One and Horizon Two – precisely because it is the perceptual substrate that makes world models possible. But crossing into Horizon Two requires the additional causal reasoning architecture on top.

Together they cross the threshold from correlation to causation. From seeing to knowing.

So we add one more piece to Will’s essay – the critical causal element of PI.

Causal Planetary Intelligence

Will ends his essay reaching for something he doesn’t quite name. Giving AI planetary sensing is, he argues, an act of embodiment, the same developmental condition that allows a child to build a causal model of the world.

He is right. But he stops exactly where the most important question begins.

What happens when you close that loop at planetary scale? That is what I call Causal Planetary Intelligence or Causal PI.

Causal PI is the completion of Will’s vision, not a departure from it. Where PI delivers situational awareness – a continuously updated picture of what is happening – Causal PI delivers simulated foresight. The ability to reason about why things are happening, model consequences of interventions before they are made, and surface decision paths most likely to produce the outcomes that matter.

The architecture has three layers.

The first is Will’s; that is planetary sensing at continuous global scale. The LEM layer. Perception grounded in physical reality.
The second is the world model layer. These are the domain-specific causal engines that represent not just physical states but the mechanisms connecting them. A wildfire causal engine encodes relationships between fuel load, wind, humidity and topography, not as statistical patterns but as causal structures that can be simulated forward and interrogated. These are not general models. They are deep, domain-specific representations of how particular physical systems actually work.
The third is the human layer. This is where Causal PI diverges sharply from the autonomous AI narrative. The human doesn’t disappear, we are indeed elevated. At Stage 4 the human is the decision validator. The causal engine runs the scenarios. The human decides with a clarity no dashboard could previously provide. At Stage 5 something more profound happens. The system becomes a second brain in the room, continuously simulating thousands of decision paths, surfacing options, reasoning over consequences, while the human operates as strategic authority. Not replaced. Not merely assisted. Genuinely augmented. This is what I term Augmented Planetary Intelligence.

Returning to Will’s fire chief. In his framing she receives a comprehensive situational picture; where the fire is, how fast it is moving, which communities are at risk. Extraordinary compared to what came before. But it is still fundamentally a better map. She still needs to make the causal leaps herself.

With Causal PI she doesn’t receive a map. She receives simulated decision paths. The causal engine has already run the wind-shift scenario, modelled the corridor failure, ranked evacuation sequences by predicted outcomes. She is not processing information. She is validating decisions the machine has already reasoned through, and making the final call with a quality of understanding no previous system could give her.

This is the elevation Will gestures at. The human moves from processor to strategic authority not because the machine has taken over, but because the machine has done the causal heavy lifting

Geography always promised causal explanation. Geospatial delivered correlation. LLMs compounded it. LEMs broke through to perception. World models introduce causal reasoning. Causal Planetary Intelligence closes the loop — sensing, understanding, simulating, deciding, at the scale of the entire Earth.

That is Horizon Two.

Will is building the sensing architecture. The causal reasoning layer I’ve described here is the natural next problem. I’d welcome the chance to think through it together. I suspect our perspectives are more complementary than either of us have fully mapped.

Matt Sheehan

June 1, 2026

Seeing Is Not Understanding. Why the Causality Gap Is AI’s Next Frontier.

TL;DR: AI systems like Planetary Intelligence can now perceive the world at scale — but perception isn’t understanding. The real frontier is causal reasoning: knowing not just what is happening, but why, and what changes if you intervene. From self-driving cars to insurance to dashboards, the same gap keeps appearing — more data and better visualization don’t close it. The next leap requires AI that can reason about cause and effect, not just correlate patterns.

1. Planetary Intelligence Can See. It Cannot Yet Understand. Matt Sheehan, LinkedIn

This week’s scan is anchored by my own piece, published earlier this week and the conversation that prompted everything that follows. Will Marshall’s vision of Planetary Intelligence is the most compelling articulation yet of where planetary-scale AI is heading — continuous sensing, Large Earth Models, edge computing, closing the decision loop faster than any human system ever has.

But there is one critical piece missing from the architecture. Planetary Intelligence gives AI the ability to perceive the physical world. What it doesn’t yet give it is the ability to reason causally — to understand not just what is happening, but why, and what would happen differently if you intervened. That gap is the thread running through every article in this week’s scan.

Read the article →

2. Why World Models Must Do More Than Simulate Pony.ai CTO, KR Asia

The CTO of Pony.ai argues that the self-driving industry made a fundamental error — assuming that more data, more compute, and better simulation would be sufficient. His point cuts to the heart of the correlation-causation divide: a world model that can only generate scenarios is not enough. It must represent how the world actually works, model interactions causally, and — crucially — be able to identify where its own assumptions are failing. He calls this diagnosability. The system must know what it doesn’t know.

This is the same architectural gap I describe in this week’s pieces. Perception is not understanding. Seeing is not knowing. And in high-stakes, real-time decisions — whether a robotaxi or a fire chief — the difference is everything.

Read the article →

3. From Raw Data to Smarter Decisions: Decision Intelligence Best Practices Forbes Tech Council

A practitioner-oriented guide that maps Decision Intelligence as a formal discipline — not a technology, but an architecture for how decisions are made, evaluated, and improved over time. The key finding that should give every data leader pause: more than a quarter of data and analytics teams estimate annual losses above $5M from poor decision architecture. Seven percent put that number above $25M.

The piece is grounded in industrial operations, but the principle is universal. The problem isn’t a lack of data. It’s the absence of a structured loop connecting data, human expertise, and decision outcomes. That loop — continuous, causally informed, improving — is exactly what Horizon Two requires.

Read the article →

4. When Risk Moves Faster Than Insurance, Everyone Pays InsuranceNewsNet

This piece is about the insurance industry, but read it as a systems failure story. Climate risk is now moving faster than the pricing models designed to contain it. Carriers built on historical correlation — past loss patterns, actuarial averages, static risk zones — are finding those models increasingly blind to a world where the causal dynamics are shifting in real time.

This is the fire chief problem at industry scale. Better historical data doesn’t help when the underlying causal structure of the system has changed. What’s needed isn’t more correlation. It’s models that can reason about why risk is moving, simulate forward, and price accordingly. The article doesn’t use that language — but that’s exactly the gap it’s describing.

Read the article →

5. The Illusion of Control: Why Dashboards Are Failing Legal and Operations Teams Forbes Tech Council

Dashboards were built for summarisation. They tell you what happened. They almost never tell you why it happened, and they cannot tell you what to do next. This Forbes piece makes that argument for legal and operations teams — but the diagnosis applies everywhere.

The illusion of control is precisely what I mean when I describe the decision bottleneck that Planetary Intelligence, in its current form, doesn’t yet solve. A better dashboard is still a better map. The fire chief still has to make the causal leaps herself. Until the architecture shifts from reporting the past to simulating the future, the bottleneck remains — regardless of how sophisticated the visualisation becomes.

Read the article →

Matt Sheehan

May 27, 2026May 27, 2026

The Geospatial Gap Is Widening. Four Signals You Can’t Ignore

TL;DR: The world model bet just got $450M more serious. Google is already building planetary-scale geospatial reasoning without the industry’s help. And most geospatial organisations are still trying to sell the future with a 2019 business model. This week’s four reads explain why that gap is about to become a chasm.

Matt Sheehan | Spatial-Next: Geospatial Is Looking for Answers in All the Wrong Places

The geospatial industry is in the middle of its most significant transformation in fifty years — and most organisations are navigating it with the wrong map. In this week’s feature article, Matt argues that restructuring data for machine consumption is not the destination, it is merely the entry ticket. The harder problem is the Commercial Architecture Trap: organisations that have built genuinely Horizon Two products but are still trying to sell them with Horizon One pricing, sales motions, and success metrics. The result is a product that has crossed the line, trapped inside a business that never followed. Read the full article on the Spatial-Next blog.

Read the article: https://www.linkedin.com/pulse/geospatial-looking-answers-all-wrong-places-matt-sheehan-byv1c/

Demis Hassabis | DeepMind: Language Models Can’t Understand Reality

Think of the most capable AI tool you have ever used. Now ask it to predict what happens when you push a glass off a table. It can describe the shattering in poetic detail. It has absolutely no idea why the glass falls. That is the argument DeepMind CEO Demis Hassabis laid out in a widely reported interview this year — and it is one of the clearest articulations of why world models are not a incremental step beyond LLMs, but a fundamentally different architecture. Language describes the world. It does not contain it. For anyone trying to understand where the next wave of AI is actually headed, this is essential reading.

Read the article: https://cryptobriefing.com/deepmind-hassabis-world-models-llm-limits/

Decart | $300M and the Race to Build Physical AI Infrastructure

When a two-year-old company raises $300 million at a $4 billion valuation — backed by NVIDIA, Sequoia, Toyota, and OpenAI co-founder Andrej Karpathy — it is worth asking what they are building and why the smartest money in AI wants a piece of it. Decart’s thesis is straightforward and significant: language models operate in text and don’t understand how the physical world behaves. World models are the missing layer. Their Oasis product is a world model built specifically for physical AI, and their infrastructure stack is designed to run it in real time. This is the capital signal that the shift from language to physical reasoning is no longer a research conversation.

Read the article: https://thenextweb.com/news/decart-300-million-radical-ventures-world-models

Google Research | Earth AI: Foundation Models and Geospatial Reasoning at Planetary Scale

Google quietly published one of the most consequential pieces of geospatial AI research of the year — and most of the industry missed it. Their Earth AI system pairs a family of foundation models across imagery, population dynamics, and environment with a Gemini-powered geospatial reasoning agent that can deconstruct complex real-world queries into multi-step plans and execute them at planetary scale. In a live demonstration, the agent reasoned across hurricane forecasts, population vulnerability data, and satellite imagery — training a model on the fly — to identify at-risk communities before a storm made landfall. This is Stage 4 in production. The question for the geospatial industry is not whether this is impressive. It is whether your data is structured to feed it.

Read the article: https://research.google/blog/google-earth-ai-unlocking-geospatial-insights-with-foundation-models-and-cross-modal-reasoning/

Matt Sheehan

Matt Sheehan is a senior executive and AI strategist with over 25 years of experience leading complex organizations through technology-driven transformation. He specializes in moving AI from experimentation to enterprise-scale value — closing the gap between technology capability and measurable business outcomes. Matt has built the frameworks, the teams, and the delivery systems that turn AI pilots into scalable operating models, with a particular focus on decision velocity: compressing the distance between insight and action in environments where the cost of being slow is real.

May 27, 2026May 27, 2026

The Map Is No Longer the Destination: Geospatial’s Defining Moment

TL;DR

The geospatial industry is building faster, slicker tools (AI-powered maps, natural language queries, instant analysis) — but risks missing the bigger shift happening underneath it.

It has the best fuel for the next AI engine, but is still optimizing the horse.

The author’s core argument: most geospatial AI is still pattern-matching on the past (Stage 3), while the next leap (Stage 4) is about causal reasoning and simulation — systems that understand why something will happen, not just that it looks like something that happened before.

Google I/O just showed the Stage 4 infrastructure is already being built, at massive scale, and largely without geospatial data — because that data is still formatted for humans (dashboards, maps) rather than machines.

The opportunity: geospatial companies sit on exactly the ground-truth physical data that Stage 4 world models need. But to matter, they need to restructure it for machine consumption — encoding physics, constraints, and causal relationships — not just make prettier dashboards faster.

I was reading about Javier de la Torre‘s impressive keynote at SDSC London this week. It raised in me both admiration, and a growing sense of urgency.

Javier demo’d: Live map creation, spatial analysis, app generation, agents building on top of maps, this end-to-end acceleration would have seemed extraordinary two years ago. Javier stated:

Geospatial is about to change as much as software development has changed in the last six months.

He’s right that the friction is reducing. He’s right that the acceleration is real. But as I reflect, there is one question which keeps coming to my mind ..

Are we building a faster horse, or the automobile?

Where the Industry Actually Is

To answer that question honestly, we need a map of where the geospatial industry sits on the AI maturity curve.

I have spent the last several years developing a Six-Stage AI Maturity Model that traces the journey from analog silos to anticipatory intelligence. It is not a theoretical framework. It is something I have built, broken, rebuilt, and used in complex, data-intensive environments where the cost of getting it wrong is very real.

The AI Maturity stages run as follows.

Stage 0 is the analog and silo phase: Manual retrieval, tribal knowledge, disconnected records. Data exists but is trapped. Decision latency is measured in days.

Stage 1 is overcoming blindness: Centralised dashboards, a single source of truth, the GIS layer that gives teams a shared picture of the playing field. This felt like the destination for a long time. For many organisations, it still does.

Stage 2 is the first AI phase: Solving the mechanical work problem. Natural language queries, automated extraction, the “query-to-insight” loop. An analyst asks: are there any vehicles in a high-risk zone near the water? The AI finds the answer instantly. This is where Carto demo’s live. Genuinely valuable, but not the end state.

Stage 3 is the second AI phase: Moving from reactive to proactive. The system stops waiting to be asked and starts pushing alerts. Pattern recognition across live data streams and historical archives. The AI notices that this location, this temperature, this vehicle weight combination looks like a past disaster, and flags it before the human thinks to check. This is correlation, not causation. The system knows that a risk exists. It does not yet understand why.

A critical note here: Stage 3 is where the first genuinely agentic implementations emerge — systems that can observe, reason, and escalate. Sending an emergency alert when a car enters a river is Stage 3 agentic behaviour. Impressive, and increasingly table stakes. But it is still Horizon One thinking. The system is still a sophisticated pattern-matcher — faster than any human, tireless, and operating at scale, but fundamentally anchored to the past. It can only flag what it has seen before, in some form. It has no model of the world. It cannot simulate. It cannot reason about a scenario it has never encountered. And that ceiling matters enormously, because the crises that cause the most damage are rarely the ones that look like the last one.

The leap from Stage 3 to Stage 4 is not an incremental improvement in accuracy or speed. It is a shift in the nature of the question the system can answer — from what does this look like? to what will this become? That requires a different architecture, a different data philosophy, and a different relationship between human and machine. It is the difference between a system that warns you the bridge looks icy and one that hands the decision-maker a calculated answer: this vehicle, at this speed, on this grade, in these conditions, has a 23% probability of making it across. The human still decides. They just decide with something the previous system could never produce.

Moving from Stage 3 to Stage 4 is the (giant) jump from Horizon One to Horizon Two. This is the paradigm shift.

Stage 4: Here we enter the causal AI phase – world models, simulation capability, the ability to reason about what will happen rather than what has happened before. The analyst no longer asks what the risk is. They run a simulation: if the temperature drops two degrees and the vehicle weighs 4,000 pounds, does the tire friction hold on this grade? The world model calculates the physical interaction of ice, mass, and gravity. It does not pattern-match against history. It understands physics. This is where LLMs step back from the reasoning layer and become the interface — the human-facing layer over a fundamentally different computational engine underneath.

Stage 5 is anticipatory intelligence. The frontier most organisations cannot yet see clearly. The reasoning engine simulates thousands of decision paths and proactively suggests the next best move before the crisis manifests. The human moves from data processor to governor, authorising machine-generated strategies rather than constructing them from scratch. This is decision velocity in its fullest form.

The Horizon Problem

I have written before about the two-horizon challenge facing geospatial incumbents.

Horizon One is your current business – existing customers, real revenue, proven infrastructure. It needs to be protected and maintained. Carto’s demo’s are a superb Horizon One acceleration. Taking analysts from hours to minutes, from complex workflows to natural language queries, that is enormously valuable within the existing paradigm.

Horizon Two is the fundamentally different architecture emerging to replace it. And the key word is fundamentally. Not an upgrade. Not an extension. A different value chain entirely.

The old geospatial value chain runs like this:

Data Collection → Processing → Analysis → Visualization → Human Insight → Decision

Every link optimised for a human looking at a screen and understanding something. The map, the dashboard, the analyst — these are not incidental features of the industry. They are its entire architecture.

The second horizon inverts this:

Decision Requirement → Agent → Data Query → Reasoning → Action or Human Authorization

Visualisation shifts on the critical path, but it does not leave it. The human is still in the loop: governing, authorising, deciding. What changes is its purpose. In Horizon One, the map is the product. The insight lives in the visual. In Horizon Two, the map becomes the governance layer or the interface through which a human oversees what the machine has already reasoned.

That is a profound difference in where value is created, even if the dashboard still exists.

The machine does not need a choropleth map to do its work. It needs clean, queryable, machine-addressable spatial data. GeoParquet and STAC are not faster cart wheels. They are the pipelines and refineries that make machine-speed reasoning possible, reasoning that gets handed to a human decision-maker in a form they can act on, not in a form that looks good in a boardroom.

This is the gap between what Carto demonstrated and where the market is heading. The demo’s were beautiful. The question is not whether beauty matters, it certainly does since humans still need clear interfaces to oversee, interpret, and decide.

The question is: are organisations mistaking the interface for the infrastructure?

What Google Just Told Us

The day after SDSC London, Google I/O made the urgency of this question impossible to ignore.

The numbers alone are amazing: token processing up 7x year-on-year to 3.2 quadrillion per month, $190 billion in capital expenditure this year, 8.5 million developers building on their models monthly – see the article linked to in the References section below.

But the numbers are not the story. The story is the architecture Google is building.

Gemini Spark – a 24/7 personal agent running on dedicated virtual machines, performing long-horizon tasks in the background, integrating with tools through MCP, operating directly within Chrome. Information agents in Search running continuously, personalised, finding what you need at exactly the right moment. Antigravity 2.0, an agent orchestration platform that turns the development environment into a platform for cohorts of autonomous agents.

Google is not building a faster horse. They are building the Stage 4 and Stage 5 infrastructure described in my maturity model above, and they are building it at a scale that no geospatial organisation can match.

Critically, they are building it without the geospatial industry.

This is not by choice, but by default. Because the geospatial industry’s data is largely still formatted for human consumption. Visualisation-first. Dashboard-optimised. Designed for an analyst looking at a screen, not for an agent pulling what it needs at machine speed to serve a decision requirement.

Gemini 3.5 Flash, which was launched the same day, achieves 83.6% on MCP Atlas, the agentic tool-use benchmark. It runs at 289 tokens per second, four times faster than competing frontier models. It is not pattern-matching in the Stage 3 sense. At the frontier of LLM capability, something different is beginning to emerge — the ability to reason about relationships, not just recognise patterns.

The window between Stage 3 LLMs and Stage 4 world models is compressing faster than most in the geospatial industry realise.

The Causality Gap Is the Strategic Gap

Here is the thing that keeps me up at night, thinking about where geospatial sits right now.

The industry has spent decades building extraordinary data assets; deep, clean, structured spatial data at a scale that the pure-play AI world consistently underestimates. Point clouds, digital twins, high-resolution built environment data, satellite imagery at minutes-level revisit rates, AIS vessel streams, terrain models. This is not legacy liability. This is the exact fuel that Stage 4 world models need.

The question is not whether geospatial has value in the second horizon. It absolutely does. The question is whether the industry restructures that data for machine consumption before someone else builds the engine without it.

Because here is what Stage 4 actually requires that the geospatial industry is uniquely positioned to provide: ground truth. Real-world physical constraints. The data that lets a causal model know that a truck cannot make a 35-degree turn at speed on a wet surface, that ice forms at this gradient before it forms at that one, that this building’s foundations interact with that soil type in this specific way under these load conditions.

LLMs at Stage 3 pattern-match against history. World models at Stage 4 encode physics. Geospatial data is physics … if it is formatted as such.

GeoParquet is a start. STAC is a start. But the deeper restructuring required is ontological, not just technical. The data needs to encode relationships, constraints, and causal dependencies and not just geometries and attributes designed for a human to interpret visually.

The Second Brain in the Room

In my comment on Javier’s post, I used the phrase that I keep returning to: the agent as a second brain in the room.

This is not a metaphor for replacement. I must emphasise, the goal of the second horizon is not autonomous execution. It is better decisions, made faster, by humans with machines that can actually reason. It is augmentation. The human moves upstream, from processor to governor. They set parameters, define thresholds, authorize actions. The agent synthesises data, surfaces options, and reasons over context at a speed and scale no analyst can match.

For lower-order, high-frequency decisions, examples include monitoring, alerting, routine classification, the agent acts autonomously at machine speed. For higher-order decisions, the agent becomes the second brain: another perspective in the room, working through the implications faster than any human could, presenting options rather than dictating outcomes.

This distinction matters enormously for geospatial. The industry’s concern about AI replacing analysts misses the point. The analysts who survive and thrive in the second horizon are not the ones who know how to navigate dashboards. They are the ones who understand the physical and operational constraints deeply enough to govern the systems reasoning over them.

Domain expertise does not become less valuable in Stage 4 and Stage 5. It becomes the most valuable thing in the room, and that is because it is that knowledge upon which the causal model is built on.

The Faster Horse Trap

Javier’s demos will be celebrated across the geospatial industry this week, and they deserve to be. The work is real, the capability is genuine, and the friction reduction is meaningful for organisations still operating at Stage 2.

But the faster horse trap is not about intent. It is about timing and framing.

Nokia understood smartphones were coming. They had the prototypes. Their entire organisation — culture, incentives, engineering — was built around the first horizon. Kodak invented the digital camera in their own labs in 1975 and buried it because it threatened film.

The geospatial industry is not Kodak. But it is at a Kodak moment.

The organisations that define the next decade of spatial intelligence will not be the ones who built the best dashboards. They will be the ones who recognised, early enough, that the data underneath the map was always the asset, and restructured it accordingly.

The horse didn’t lose because it got slower. It lost because the world it was built for stopped being the world that mattered.

The geospatial industry has spent decades building extraordinary data assets. The second horizon does not make that work obsolete. It makes it the most valuable fuel in the room; if the industry is willing to reformulate it for the engine that is already being built.

Google just told us the engine is ready. The only question is who supplies the fuel.

So, I believe the geospatial industry holds the most valuable data assets for the machine-speed decision layer. But data and fuel are not the same thing, at least not yet.

Is the industry ready to make that transition, or are we still celebrating the faster horse – what do you think?

Matt Sheehan

References

https://blog.google/intl/en-africa/products/explore-get-answers/sundar-pichai-io-2026/#more-news

https://www.linkedin.com/posts/jatorre_just-finished-my-keynote-at-sdsc-london-2026-ugcPost-7460626616562561026-SYxj

May 19, 2026May 19, 2026

Geospatial’s Faster Horse Problem

TL;DR: The AI revolution is creating two worlds — your existing business (Horizon 1) and a fundamentally different future (Horizon 2). Most companies are treating Horizon 2 as just an upgrade to what they already do, which is a mistake.

Incumbents have resources and relationships but risk getting stuck protecting the past. Startups can build for the future but risk running out of time and money before the market catches up.

The real killer for both? Starting with the technology instead of the problem. The winners will be those disciplined enough to lead with customer need, build for scale from day one, and — for incumbents especially — ring-fence their Horizon 2 efforts so the old business doesn’t slowly strangle them.

In a recent article: You Know AI is Changing Everything: So Why Aren’t Your Projects Scaling? I introduced the idea of the two horizons:

Horizon one is your current business – existing customers, real revenue, proven infrastructure. It’s what keeps the lights on today and needs to be protected and maintained.

Horizon two is the fundamentally different paradigm emerging to eventually replace it. The key word is fundamentally – it’s not an upgrade or extension of Horizon 1, it is a different architecture, different value proposition, different way of operating.

In that article I argued that many of today’s organizations are treating horizon two as an extension of horizon one. In other words an add-on. I used the analogy of a faster horse in: The Geospatial Value Chain is about to Invert. GeoAI, in its current form, can be seen as a good example of this. Put in the hands of an analyst new machine learning models – extracting data from an image – is a popular one.

But the move from horizon one to horizon two is not about a faster horse, it is about a new mode of transportation. In the early 1990’s that meant the automobile.

Today’s Paradigm Shift

So, as with any paradigm shift, we have new pure play horizon two players and the horizon one incumbents. The former are the start-up crowd, nimble, light-weight, innovative but fragile. The latter – slower moving, inertia-bound, established but threatened.

The technology paradigm shift of today, powered by AI, will have its winners and losers. My goal in this article is to explore how this plays out specifically in the geospatial industry: the new horizon two players, where the incumbents are vulnerable, and what the path forward looks like for those willing to make the leap.

New Waters Ahead

Simon Sinek famously argues that most organizations communicate from the outside in. In other words they lead with what they do, then how they do it, and rarely get to why. The most inspiring ones, he observed, do the opposite.

They start with why.

But when it comes to navigating a paradigm shift, I’d argue there’s a different problem. Most organizations already know the why, that AI is reshaping everything. And that window is closing .. standing still is not an option. Many are even clear on the what – a new architecture, new products, new ways of delivering value.

But it is the how that kills them.

Horizon two is a green field, in other words there are no incumbents. It is also ill-defined and vague. But it is our inevitable future.

So how will the winners win?

The How of Horizon Two

The battle for horizon two will be won or lost on execution, and the combatants are not evenly matched.

The Incumbents

For established geospatial organizations, the answer is not another R&D skunkworks that gets quietly defunded when Q3 numbers disappoint. It is a dedicated horizon two business unit – separately resourced, separately led, and protected from the gravitational pull of the existing business. This is harder than it sounds. The leadership challenge is real: how do you justify investment in something that, by definition, threatens your current revenue streams and won’t show returns on the timelines your board expects?

The ones that get this right treat horizon two not as a cost center but as a strategic bet; funded like a startup, held to different metrics, and given the organizational air cover to operate differently. The ones that get it wrong fold it back into the mothership, where horizon one culture, incentives, and quarterly pressure slowly suffocate it.

The Start-ups

The new players have the opposite problem. Unburdened by legacy architecture or existing customer commitments, they can build natively for the horizon two world: autonomous, embedded, decision-velocity first. But nimble and fragile are two sides of the same coin. Without the internal processes, distribution, customer relationships, or the domain credibility that incumbents have spent decades building, many will either build the wrong thing, or the right thing and still lose.

The winners in this category will be the ones who find a way to anchor early — a marquee customer, a strategic partnership, a niche where their horizon two architecture delivers an outcome no incumbent can match — and use that foothold to scale before the incumbents catch up.

But there is a danger that cuts across both camps equally. The temptation — for incumbents and start-ups alike — is to let this become a technology exercise, led by technologists, measured by technology metrics. It is a trap I have watched organizations fall into repeatedly, and not just in the AI era. Throughout my career, the pattern has been consistent: the projects that fail are rarely the ones that built the wrong thing technically.

They are the ones that never started with the right problem.

That means beginning with the customer, the commercial reality, and the outcome that actually matters, and working backwards to the technology. Most AI failures today trace back to this same 2 root causes:

A solution looking for a problem, rather than a problem demanding a solution.
A one off narrowly focused solution which cannot be scaled.

Avoiding these traps requires deliberately putting commercial and customer voices at the table from day one – not as an afterthought once the technology is built. And building in scaling gates as a solution evolves (more on this latter to come in future articles).

So Who Wins?

In reality – neither, automatically. The incumbents have the relationships and the resources but risk optimizing themselves into irrelevance. The start-ups have the architecture but risk running out of runway before the market is ready. What determines the outcome is not technology … it is discipline.

The discipline to run two horizons without letting one kill the other. The discipline to start with the business problem, not the tool. And the discipline to ask the hard questions early: does this solve something real, can it scale, and will people actually use it?

That is not a technology problem. It is a leadership and execution problem. And in my experience, that is exactly where most organizations – on both sides – come up short.

In my next article I will discuss what I see as the biggest challenge AI can help organizations overcome, that of decision velocity. This is a game-changing opportunity for geospatial to break out of its established niche, and past today’s actionable insight deliverable (bottleneck) – to become truly foundational to all enterprises.

Matt Sheehan

May 19, 2026May 19, 2026

The Machines Are Learning How the World Works. Are You?

The Decision Layer Signal Scan — Week of May 19, 2026

TL;DR

AI is reshaping not just what organizations do, but how they’re built, what they understand, where they get their data, and what role humans play in it all. This week: ditch the org chart, watch world models, notice who’s sitting on valuable training data — and don’t forget that judgment is still yours.

1. Forbes: Why Org Charts Are Now Obsolete The Gist: AI-native organizations can’t be built on traditional hierarchies. The argument is that org charts were designed to manage information flow and control — but AI is collapsing both of those functions. The new design principle is building around decisions and outcomes, with fluid teams, not fixed functions. Why it’s Relevant: For the Decision Layer, this is foundational. If the org chart is the skeleton of how decisions get made, and AI is reshaping decision-making itself, then the structure has to change too. Leaders who bolt AI onto existing hierarchies will find the hierarchy wins — and slows everything down.

Source: Designing an AI-Native Organization: Why Org Charts Are Obsolete (Forbes)

2. MIT Technology Review: World Models — The Next Frontier in AI The Gist: Large language models learned from text. World models are what comes next — AI systems designed to understand how the physical world actually works: motion, space, causality, consequence. MIT Tech Review flags this as one of the most significant developments in AI right now, with Google DeepMind, Fei-Fei Li’s World Labs, and Yann LeCun all racing to get there first. Why it’s Relevant: For geospatial and remote sensing, this is a signal worth watching closely. World models need spatial intelligence at their core. The industry that has spent decades building the tools to capture, model and interpret physical environments is sitting closer to this frontier than it might realise.

Source: World Models: 10 Things That Matter in AI Right Now (MIT Technology Review)

3. TechCrunch: Origin Lab Raises $8M to Feed World Models with Game Data The Gist: Origin Lab is building a marketplace connecting video game companies with AI labs that need training data for world models. The premise: games already contain the physics, movement, spatial logic and cause-and-effect dynamics that world models need — and no one had built the infrastructure to make that data accessible and licensable at scale. Backed by Lightspeed, with angels including Twitch co-founder Kevin Lin and Cruise founder Kyle Vogt. Why it’s Relevant: An unexpected data source becomes a strategic asset. The parallel for geospatial is direct — decades of aerial, satellite and sensor data represent exactly this kind of rich, structured, physics-consistent environment data. The question is who builds the infrastructure to make it flow to where it’s needed next.

Source: Origin Lab Raises $8M to Help Video Game Companies Sell Data to World-Model Builders (TechCrunch)

4. DisrupTV Ep. 439: The Human Edge in an Age of Agentic AI The Gist: Vint Cerf, Dr. David Bray and Cheryl Strauss Einhorn — joined by hosts Ray Wang and Vala Afshar — dig into what remains distinctly human as AI agents take on more of the decision-making load. Einhorn’s framing is sharp: AI generates answers, but it doesn’t know you and it doesn’t care about consequences. You do. Bray adds a useful reframe — stop thinking “human-in-the-loop” and start thinking “AI-in-the-group.” Why it’s Relevant: As agentic AI moves from concept to operational reality, the human edge isn’t about resisting automation — it’s about bringing wisdom, accountability and better questions to the table. For anyone leading AI transformation, this is a grounding conversation worth an hour of your time.

Source: The Human Edge in an Age of Agentic AI — DisrupTV Ep. 439 (Constellation Research)

Matt Sheehan

May 13, 2026May 13, 2026

Your Strategy is Only as Fast as Your Slowest Decision

1. Matt Sheehan | Spatial-Next: You Know AI is Changing Everything: So Why Aren’t Your Projects Scaling?

The Gist: Most AI projects fail not because the technology is wrong, but because they are technology-led rather than outcome-led. The organisations scaling AI successfully started with a business problem. Two frameworks — the Six-Stage AI Maturity Model and the Opportunity-to-Value Framework — give leaders both a map of where they are and an operating system to get there.

Why it’s Relevant: This is the execution problem at the heart of the Decision Layer. Knowing AI matters is not enough. The discipline to ask three hard questions before any project moves forward — does this solve a real problem, can it scale, and will people actually use it — is what separates pilots from platforms.

Source: You Know AI is Changing Everything: So Why Aren’t Your Projects Scaling? (Spatial-Next)

2. Matt Sheehan | Spatial-Next: The Geospatial Value Chain is About to Invert

The Gist: The geospatial industry has spent decades optimising for human consumption — maps, dashboards, analysts. But AI can now consume and act on spatial data at machine speed, making the human-in-the-loop the bottleneck. The winners will reformat their data for machines to query and reason over directly. The losers will respond to this shift by building a better dashboard.

Why it’s Relevant: This is a precise, industry-specific illustration of what a second horizon looks like in practice. The value chain doesn’t just speed up — it structurally inverts. Data stops being raw material processed for human insight and becomes the fuel for autonomous, machine-speed decision making. Every industry with a legacy data delivery model should be asking the same question.

Source: The Geospatial Value Chain is About to Invert (Spatial-Next)

3. Insurance Edge: Stand World Model Makes its Debut

The Gist: Stand has launched what it describes as the first physics-native frontier model for the built environment — simulating how fire, wind, water, and seismic forces interact with individual structures at sub-meter resolution. In a validation against the January 2025 California wildfires, it correctly predicted structural survival outcomes at nearly double the accuracy of traditional insurance models.

Why it’s Relevant: This is a real-world horizon two player in action. Stand isn’t building a better risk dashboard — it is restructuring the entire decision architecture around physical truth computed at machine speed. It is also a compelling example of starting with the problem: a $1.3 trillion coverage gap in California, solved not by pricing risk more accurately but by designing it out, structure by structure.

Source: Stand World Model Makes its Debut (Insurance Edge)

4. Forbes Technology Council: Hidden Supply Chain Factors That Can Derail Business Strategy

The Gist: Technology strategy looks strong in theory until supply chain realities intervene. From sub-tier supplier concentration and hardware availability to decision latency and AI agent autonomy, Forbes Technology Council members identify the hidden factors — often invisible in the boardroom — that determine whether a strategy actually executes.

Why it’s Relevant: A timely reminder that decision velocity has an upstream problem. Even the best AI architecture stalls if the data feeding it is disconnected, delayed, or locked inside fragmented systems. As one contributor puts it: value is not created by insight, it is created by how quickly you turn that insight into action. That is the Decision Layer problem in a single sentence.

Source: Hidden Supply Chain Factors That Can Derail Business Strategy (Forbes)

May 13, 2026May 13, 2026

The Geospatial Value Chain is about to Invert

TL;DR: The geospatial industry is still building “faster horses” — better dashboards and analysts — when the game has fundamentally changed. AI can now consume and act on spatial data at machine speed, making the human-in-the-loop the bottleneck. The winners will be companies that reformat their data for machines to query and reason over directly, not for humans to look at. The losers will be those who respond to this shift by building a better dashboard.

There was a famous discussion at the time of the arrival of the automobile – early 1900’s – that if you asked people what they most wanted they would have said a faster horse. We are in a similar conversation today, but instead of a horse the traditional geospatial industry has focus on a faster analyst.

I discussed this in part in my last article: You Know AI is Changing Everything: So Why Aren’t Your Projects Scaling? In this article I covered the first and second horizon’s respectively. The first horizon is today’s paradigm; that is data processes and delivery to human’s via maps and dashboards. This is the promise of ‘actionable insight’, as industry players are keen to market. Today that has become a critical bottleneck to decision making.

The combination of AI and the tsunami of new data has opened the second horizon, or a paradigm shift. This will dramatically impact decision velocity.

And it is this which will turn the geospatial value chain on its head.

Today’s Geospatial Value Chain

Geospatial has attached its value to maps and dashboards; the human interface to insight. Data is that necessary foundation, but it has always been treated as a means to an end: raw material to be processed, interpreted, and ultimately translated for a human decision-maker.

The traditional chain looks something like this:

Data Collection → Processing → Analysis → Visualization → Human Insight → Decision

Every link in that chain has been optimized for the final consumer: a person. Projections chosen for legibility. Symbology tuned for interpretation. Dashboards built for executive comprehension. The entire architecture of the industry – its tools, its talent, its business models – has been engineered around the moment a human looks at a screen and understands something.

This made sense when humans were the only things capable of consuming geospatial information. And for decades, getting that chain faster, cheaper, and clearer was the competitive frontier. A better analyst. A slicker dashboard. A faster render.

Faster horses.

But here is the problem. That final link – human interpretation – is now the slowest thing in the chain by several orders of magnitude. Data volumes have exploded. Satellite revisit rates have collapsed from days to hours to minutes. Sensor networks generate continuous streams. AI models can now process, correlate, and act on spatial information at machine speed.

And yet the industry’s answer has largely been to hire more analysts, build better dashboards, and add more layers to the map.

That is not a technology problem. It is an architectural one. The value chain was built for a world where humans were the only available processor.

And that world is ending.

Inverting The Geospatial Value Chain

So when I say the geospatial chain is inverting, am I simply revisiting the old – “data is the new oil” trope?

Not at all, I simply mean the definition of usable data has changed.

That might not seem like a big deal, but it is huge. The second horizon is not about better human workflows; it is about new machine readable data forming the connective tissue of a machine-readable Earth.

In other words the consumption layer is changing entirely. Format, latency, schema, ontology; these matter more than visualization, cartography, or even interpretability. GeoParquet and STAC aren’t faster cart wheels – they are the pipelines and refineries that make machine-speed decisions possible. Opus 4.7 is the engine that knows what to do with the fuel; more on this in a moment.

So what does this new value chain look like?

The Second Horizon Geospatial Value Chain

That word inverting suggests we are simply reversing the arrows in the old value chain. Incorrect. We are talking here about a more fundamental restructuring.

Decision Requirement → Agent/Model → Data Query → Real-time Processing → Action

There are some major key structural differences here:

First, it starts from the decision, not the data. Rather than collecting data and seeing what insight emerges, second horizon systems begin with a defined decision requirement and pull only what is needed. The question drives the data, not the other way around.

Second, visualization disappears from the critical path. Now it does not vanish entirely – humans still need oversight and suggested decision-making interfaces – but it’s no longer in the critical path. The machine does not need a choropleth map. It needs a clean, queryable, machine-addressable dataset.

Third, analysis becomes continuous, not episodic. Traditional geospatial delivers insight in reports and dashboards, these are reactive, discrete moments. The inverted chain runs continuously, updating decisions as conditions change.

Lastly, the human moves from processor to governor. The nature of this shift depends on the decision at hand. For lower-order, high-frequency decisions – monitoring, alerting, routine classification – the agent acts autonomously, at machine speed, without waiting for human input.

For higher-order decisions, the dynamic is different. Here the agent becomes what I think of as another brain in the room: synthesising data, surfacing options, and reasoning over context at a speed and scale no analyst can match. But the human remains in the loop, augmented rather than replaced, making the final call with better information than they have ever had before.

In both cases, the human moves upstream. Defining parameters, setting thresholds, and intervening at exceptions rather than sitting inside every decision loop. The value of human judgment does not diminish in the second horizon.

So, when we consider the flow, visually the contrast is:

Old: Data → [human makes sense of it] → Decision

New: Decision requirement → [machine pulls, processes, acts] → Human as governor, not processor

This is the second horizon in a nutshell.

Revisiting the Faster Horse

Let’s go back and consider the horse replaced by automobile. Consider the fuel source of each of these transportation modes: oats versus oil. Oats had to be consumed by something with a digestive system. Oil can be processed industrially, at scale, without biological limitation. Data today is still largely being ‘grown’ as oats: formatted, delivered, and visualized for human digestion. The second horizon requires it to be refined as fuel.

Think about that.

Now this is not meant to be a technical article, but we are moving towards a new processing layer. For those techies in the room, the architecture looks like this:

GeoParquet → machine-readable, cloud-native, queryable spatial data

Opus 4.7 is Anthropic’s latest model, notable for its ability to reason over meaning and relationships in data, not just pattern-match on it. It fits into this flow as follows:

Opus 4.7 → semantic reasoning layer that understands what the data means and how to use it in service of a decision requirement

That flow looks like this;

Decision Requirement → Agent → [queries GeoParquet] → [Opus 4.7 reasons over results] → Action or Human Decision

In other words an agent can receive a decision requirement, find and query the right data at machine speed, reason over what it means, and act or become that second brain in the room for a human. This closes the loop at a speed no human analyst can match.

This is decision velocity.

Winners and Losers

In my last article I mentioned the need for geospatial incumbents to maintain their horizon one focus, but to explicitly fund and resource the second horizon separately. That is part of tomorrows winning formula, but so too is where the incumbents sit today.

Let’s close this article out by considering that.

The incumbents with the most to lose today are those whose entire value proposition sits in the visualization and delivery layer or the final mile of the old chain. If the machine no longer needs the map, the map business faces an existential question.

But that is not the whole story.

The large global mapping and location data platforms have something the pure-play AI world desperately needs and consistently underestimates: deep, clean, structured geospatial data at scale. That is not a legacy liability. That is a foundation …

If they choose to build on it rather than defend around it.

Their opportunity is in becoming the semantic data infrastructure layer for autonomous and agentic systems. They already have the data. The question is whether they restructure it for machine consumption rather than human delivery.

The opportunity is different but equally real for those sitting on dense, structured 3D spatial data: point clouds, digital twins, high-resolution built environment data. This is exactly the kind of spatial data that world models need. The question for these companies is whether they position as visualization tools or as machine-readable spatial data layers for AI systems.

The losers will be those who answer the second horizon with a better dashboard. The winners will be those who recognize that the most valuable thing they own is not their platform, it is their data, restructured as fuel.

Closing Thoughts

The horse didn’t lose because it got slower. It lost because the world it was built for stopped being the world that mattered.

The geospatial industry has spent decades building extraordinary horses. The question now is not how to make them faster. It is whether the industry has the courage to build the automobile, and the wisdom to recognise that the data it already holds is the most valuable fuel in the tank.

The second horizon is not coming. It is here. The only question is who builds it.

Matt Sheehan