It is 1:23 a.m. on February 15, 2021, and the man watching the Texas grid is very good at his job.

He has spent years training for moments like this. He knows what healthy grid frequency looks like. He knows what the numbers mean when they start moving wrong. And right now, they are moving very wrong.

Generators are tripping offline faster than he can track. Demand is outpacing supply. The grid is on its own — islanded from neighboring systems, running on what it has left.

He has four minutes and thirty-seven seconds before the cascade becomes unrecoverable. A collapse that could have lasted weeks. Tens of millions without power. Infrastructure damage that takes months to repair.

He has more data than any grid operator in history. Sensors. Monitoring systems. Analytics running in real time. A complete picture of exactly what is happening.

He has no system that can tell him what to do next.

He makes the call to shed load. The right call, made under impossible conditions, by a professional doing exactly what he was trained to do. It prevents the worst outcome.

Four million Texans still lose power for days in freezing temperatures. At least 57 people die.

The data didn’t fail. The sensing layer worked.

The architecture underneath it didn’t exist.

The layer nobody builds first

Before we dive into technology – sensors or AI – let’s talk about the person in the control room whose job it is to act.

The ERCOT grid operator in those early hours of February 15 was not an incompetent person with bad data. He was a trained professional, watching a system he understood, making decisions under conditions no training fully prepares you for; cascading failures, incomplete information, irreversible consequences, seconds to act.

What he needed wasn’t better data. He needed a system designed to help him use his judgment at that critical moment. This is the human decision layer. In critical infrastructure it is almost never designed. What we actually have is a trained professional, in a room full of screens, watching a crisis unfold.

That is not a decision layer. That is a person in a room with data.

Designing the human layer means asking different questions before the crisis arrives. What does this person need to know to exercise genuine judgment under pressure? What intervention options exist, and what are their downstream consequences? Where is the uncertainty in the system, and is it visible or buried? What does it mean to challenge what the monitoring system is telling you, and what protection exists for doing so?

Nobody asked those questions before February 15, 2021. The operator was positioned as a monitor, not a decision architect.

The layer that could have changed everything

Here is what a causal reasoning layer would have done differently.

It would not have predicted the storm, a causal layer would have simulated the cascade before it started.

A causal reasoning architecture encodes not just what is happening but why — and what happens next based on different decisions. For a power grid that means understanding the physical relationships between generation capacity, fuel supply, demand load, and temperature thresholds. N

That does not mean correlations. That is not the AI we already have. Causal structure is a different architecture entirely.

With that architecture in place, the question changes. Instead of: what is happening right now? The operator is asking: given current conditions, what happens if generator cluster A trips offline in the next two hours? What happens to natural gas supply if we cut power to these sectors? Which load shedding sequence preserves the most generation capacity? What is the probability this becomes unrecoverable?

When ERCOT ordered utilities to reduce power demand, transmission companies inadvertently cut power to parts of the natural gas supply chain, that meant gas producers couldn’t deliver enough fuel to the power plants that needed it. A causal layer would have surfaced that dependency before the decision was made. The consequence was not unforeseeable. It was unmodeled.

That is the difference between a sensing system and a reasoning system. The sensing system told operators what was happening. A causal reasoning system would have told them what might happen next based on different actions.

The ground truth that was already there

The sensing layer was not the problem.

ERCOT had grid frequency monitoring, generation capacity data, demand forecasting, weather feeds. The physical infrastructure was largely in place. The underlying data existed.

The data is not the bottleneck. The reasoning architecture on top of it is.

Layer one — sensing — is built, often at enormous cost, across utilities, emergency management, healthcare, finance. What has not been built is the layer that turns what those sensors see into causal understanding. And what has almost never been deliberately designed is the layer that connects that reasoning to the human who has to act on it.

What resilience actually requires

The Texas grid failure produced legislation, investigations, and weatherization requirements. All necessary. None of it addresses the architecture problem.

Winterizing generators reduces the probability of generation failure in extreme cold. It does not give the grid operator a reasoning system that can simulate cascade consequences before they execute. It does not design the human decision layer so that the person watching the screens has the authority, the tools, and the organizational standing to intervene based on judgment rather than protocol.

Resilience is not the absence of failure conditions. It is the capacity to reason about failure before it cascades, and to have a human decision layer designed to act on that reasoning in time.

The ERCOT operator had four minutes and thirty-seven seconds.

He needed a system that had already run the scenarios hours before he needed to act on them.

The sensing layer is built. The causal AI layer is coming. The human decision layer — the one that sits between causal AI output and human action — is the one nobody is designing. And it is the one that determines whether the other two are worth anything when it matters.

What is needed is a system designed to treat human judgment as the point, not the problem.

Matt Sheehan

Matt is a geographer and AI strategist with 25 years at the intersection of geospatial intelligence and decision-making. He maps the architecture connecting three layers most organisations haven’t yet seen together: the sensing layer the geospatial industry has built, the causal reasoning layer now arriving, and the human decision layer nobody is designing.

The third layer is where the value is. And where Matt has his primary focus.

If that layer hasn’t been designed in your organisation yet, that’s usually where the conversation starts: mattsheehan@spatialnext.io