In my last article, I wrote about a fire chief who needed a second brain, not a better map. This is what happens when the second brain has no one to answer to.

The fire chief in that piece was exceptional at her job. That was the point. This is about what happens to exceptional people when the system around them isn’t designed to use what they know.

A Medicare claims reviewer at UnitedHealth looked at a patient’s case. The AI had already decided. It said no.

He said yes. He was fired.

This wasn’t an edge case or a rogue manager. According to a federal lawsuit now working its way through the courts, UnitedHealth’s naviHealth AI platform was deployed to automate coverage decisions for Medicare Advantage patients. When patients appealed those decisions, nine out of ten were reversed. The AI was wrong at industrial scale. And employees who had stepped at the decision making stage to override AI’s recommendations, based on clinical reality, were disciplined and terminated.

The human was in the loop. His judgment was not.

That distinction is everything. And to understand why, you need to understand the architecture underneath it.

Three layers. One that nobody is building.

I’ve been mapping what I call Causal Planetary Intelligence across this article series — the architecture that connects AI perception to genuine decision-making. It has three layers.

Layer one is the sensing layer. Satellites, sensors, data feeds — a continuous high-fidelity picture of the physical world. In healthcare it’s the clinical data, the patient history, the diagnostic outputs. This layer is largely built.

Layer two is the causal layer – world model’s and similar systems that don’t just perceive what is happening but reason about why, and simulate what happens next if you act. This is where the research frontier sits right now. It is real, it is coming.

Layer three is the human decision layer. The reviewer. The clinician. The underwriter. The incident commander. The person whose judgment sits between machine output and consequential action. This is the layer the UnitedHealth case just broke in public.

The design failure nobody is naming

Every organisation deploying AI at consequential decision points will tell you the same thing: there’s a human in the loop. A reviewer. An approver. A sign-off. The human is there.

What almost nobody asks is what that human is actually authorised to do.

In the UnitedHealth case the answer was: not much. The AI set the direction. The human was positioned as a checkpoint, not a decision-maker. When he exercised independent judgment the organisation treated it as a defect rather than the feature it was designed to be.

This is not a UnitedHealth problem. It is a design problem being replicated across industries at speed as AI moves into consequential decision territory.

The sensing layer is built. The AI is running. The human is present. But the interface between machine output and human authority — who can challenge it, on what grounds, with what protection, and within what timeframe — has been left to chance. Or worse, designed backwards.

Torsten Kriening put it precisely in a comment on my last week’s fire chief article:

“The human decision layer resists a technical solution, because trust, transparency and an honest accounting of what the system does not know are organisational and design problems, not compute problems. It tends to be the least funded precisely because it is not a deliverable anyone can ship.”

The UnitedHealth case is what happens when you skip that layer entirely. You build the AI. You put a human next to it. You call it human oversight. And then you systematically remove the human’s authority to actually use it.

That is not a human decision layer. That is a human in a costume.

Legislators are noticing

When design fails at scale, regulators arrive. And they are arriving.

Colorado just signed a revised AI law — a rewrite of its original 2024 act, signed May 14, 2026 — requiring organisations to designate a specific individual, trained and authorised, who can review and potentially override consequential AI decisions. It takes effect January 1, 2027.

Maryland’s law, effective October 1, 2025, requires health insurers to report quarterly on adverse AI decisions — how many, what type, and whether AI was a factor.

Multiple states are now legislating to prohibit AI as the sole basis for insurance denials.

What legislators are trying to write into law is something that should have been designed in from the beginning. Not a checkbox. Not a human rubber stamp positioned after the AI has already decided. A genuine decision authority; someone with the training, the tools, the protection, and the organisational standing to challenge machine output when clinical or operational reality demands it.

Colorado’s law can mandate that role into existence. It cannot design what that role actually requires.

The gap nobody is talking about

Recent research on AI governance maturity found that only one third of organisations had reached a level adequate for the autonomous agents they were already deploying. Two thirds of organisations are running AI at consequential decision points without the governance architecture to support it.

The harder finding: when an AI agent makes a wrong decision autonomously, that decision has already executed.

The damage is done before a human ever sees the log.

That is the UnitedHealth case written at scale. The AI ran. The human was present but not authorised. The damage accumulated. The lawsuits followed.

What designing the human layer actually requires

The Modern War Institute published a piece earlier this year on AI in military decision-making that contained a line worth carrying beyond its original context:

“The primary locus of control is not at the moment of decision. It is at the moment of design.”

This is the insight most AI deployments are missing. They put in place decision points; add a reviewer, a sign-off, an approval workflow, and assume the human layer is handled.

It isn’t.

Designing the human layer means asking different questions before the build starts. Who is this decision for. What does the human need to know to exercise genuine judgment. What does it mean to challenge the AI’s recommendation and what protection exists for doing so. How do you make uncertainty visible rather than burying it in a confident output. How do you know the human is adding judgment rather than just absorbing liability.

These are not engineering questions. They are not compliance questions. They are design and architecture questions that belong at the beginning of every AI deployment; not at the end when the lawsuits arrive.

The UnitedHealth reviewer saw something the AI missed. He had the expertise. He had the clinical context. He had the judgment.

What he didn’t have was a system designed to use it.

That is the human layer problem. And it is available to solve right now; not when the regulatory landscape settles, but in every organisation deploying AI at key decision points today.

The fire chief doesn’t need a better map. The Medicare reviewer doesn’t need a better AI.

They both need the same thing: a system designed to treat human judgment as the point, not the problem.

Matt Sheehan

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

Most organisations have invested heavily in the first layer and are beginning to see the second arriving. Very few have asked what happens to the human when it does — who receives that reasoning, what they’re authorised to do with it, and whether the system is designed to use their judgment or route around it.

If that question hasn’t been asked in your organisation yet, that’s usually where the conversation starts. Reach Matt directly at mattsheehan@spatialnext.io