Computers cannot make decisions

The headline running through tech finance today is essentially this: Palantir is up because sovereign AI demand is real. Analysts are upgrading the stock. Nations are buying the infrastructure. The room is filling with hardware.

But the headline nobody printed is the one I keep thinking about.

Computers cannot make decisions.

---

I do not mean that as a limitation to fix. I mean it as a structural fact that leadership theory has to take seriously before the procurement contracts close.

A model can rank options. It can surface probabilities. It can tell you which of three routes costs less, which of four vendors has the shorter latency, which policy scenario has historically produced the better outcome in a comparable population. That is genuinely useful. I build with these tools every day.

But ranking is not deciding. Deciding requires someone who will carry the consequence. A model carries nothing. The tribe carries it. The county carries it. The clinic staff carries it. The family carries it.

When we confuse the ranker for the decider, we do not save time. We displace accountability. And displaced accountability has a way of landing on whoever had the least power in the original room.

What participatory sensemaking actually means

In Adaptive Inclusive Leadership Theory, one of the four core constructs is Participatory Sensemaking. The claim is precise: meaning is made collectively, and decisions made with the room hold longer than decisions made for the room.

Hold longer is doing real work in that sentence. I do not mean the process feels better or that morale improves. I mean the decision persists. It survives the first hard quarter. It survives the staff turnover. It survives the moment when the tribal council gets a new chair or the county commissioner changes.

Decisions made for the room collapse under those conditions because no one in the room owns them. They were handed down by a process, a consultant, a model, an outside authority. When the authority leaves, the decision softens.

Decisions made with the room have a different texture. The people in the room argued about it. They shaped it. They gave something up to reach it. That transaction creates ownership, and ownership creates durability.

The sovereign AI version of this problem

Here is where it connects to what is actually happening in the market right now.

Nations and communities are making infrastructure decisions at speed. The sovereign AI buildout is real. The demand is real. The vendor attention is real and intensifying. I watch it from Ada every morning.

But infrastructure decisions are also governance decisions. Who controls the node controls what the node sees, what it retains, what it shares, and who it answers to. That is not a technical question. That is a sovereignty question. And sovereignty questions require the room.

I have watched tribal technology decisions get made with the vendor in the room and the community out of it. The hardware arrives. The deployment runs. The dashboard populates. And then, somewhere between month four and month eighteen, something goes sideways because the people who live with the system were never part of making it.

The same pattern runs through rural hospital deployments, county government digitization projects, and school district AI pilots. The gap between the decision meeting and the people affected by the decision is where implementations die.

What a leader does that a model cannot

A leader convenes the room. That sounds simple. It is not.

Convening means identifying who is missing and going to get them. It means creating enough psychological safety that the person with the lowest status in the building will say the thing everyone else already knows but will not name. It means staying in the room when the ranking the model produced conflicts with what the elder or the front-desk clerk or the overnight maintenance worker knows from experience.

Models produce outputs. Leaders produce conditions. Those are different jobs.

The sovereign AI infrastructure moment we are in right now needs both. The compute is necessary. The governance is the layer that makes the compute worth having. And governance is not a setting in the software. It is a practice that lives in the people who use the system, and it has to be built the same way any durable decision gets built: with the room, not for it.

The cost of skipping it

I will name the trade-off plainly because I think it gets smoothed over too often.

Participatory sensemaking is slower at the front end. Convening the room takes calendar time. Working through disagreement takes emotional energy. The vendor who shows up with a finished slide deck and a clean deployment timeline will look more efficient than the leader who says, wait, we have not heard from the people in the outlying district yet.

That efficiency is real on day one. It is a liability by month twelve.

The cost of skipping participatory sensemaking is not a soft cost. It is rework, resistance, quiet non-adoption, and eventual system abandonment. In sovereign and rural communities where resources are already thin, that is not an abstract risk. It is a specific budget line that never appears on the original proposal.

Build the room first. Then buy the rack.

Yakoke.