Cities are learning to predict problems. The harder question is whether they can act on them.

By Eddie Cheng, Co-Founder and Chief Technology Officer of Origen

Ask whether a city can predict a problem before its residents feel it, and the honest answer in 2026 is yes. Weather models, satellite imagery and dense sensor networks can now tell a municipality which underpass will flood, which junction will lock up and which substation is running hot. Prediction is no longer the scarce capability. The real challenge is turning a prediction into a decision, and a decision into an action that physically changes what happens on the street. A forecast that cannot drive either is simply a more sophisticated way of watching things go wrong.

The pressure to close that gap is not abstract. The United Nations’ World Urbanization Prospects 2025 finds that cities are now home to 45 per cent of humanity, that nearly two-thirds of global population growth through 2050 will occur in urban areas, and that the number of megacities has quadrupled over the past five decades. At that scale, a management model that waits for a complaint before raising a work order is not merely slow; it is arithmetically outmatched. No control room will ever have enough people to respond to a city of ten million, one incident at a time.

Instrumented is not intelligent

Much of the past decade’s smart-city investments went into instrumentation: more sensors, cameras and dashboards. That work was necessary. But a sensor count is a procurement metric, not an intelligence metric. A city with 10,000 sensors and 40 dashboards is well instrumented, yet not intelligent. Intelligence begins when the data from those sensors can be reasoned over as a single picture and converted into an action that has physical impact.

Three transitions, and where most cities stall

In our work with municipalities, we describe the journey as three transitions. Most cities are still working on the first.

From invisible to visible. Most of what a city needs to know about itself already exists, but it is locked in silos: utilities in one system, construction permits in another, and resident records in PDFs and flat databases with no spatial depth. Each department monitors only its own part of the city, whether zoning, utilities or construction, while no one sees the whole picture. Bringing those records together in a single, spatially and temporally aligned three-dimensional model of the city is unglamorous work. It is also where much of the value is created, because a fact only becomes actionable once its location and timing are known.

From reactive to proactive. Once the whole scenario is understood, questions change. Rather than investigating after a storm why a district was underwater, a city can run a flood simulation on a proposed development before the first foundation is poured and change the design rather than the drainage budget.

From decision to action. This is the transition that matters most and is attempted the least. Consider utility corridors, for instance. A pipeline conflict discovered during excavation can delay a project by months. When the same conflict is detected in digital twins, stakeholders are alerted and the design change is pushed back into the engineering systems, it can be resolved before anyone lifts a shovel. The next generation of city command centres should not be judged by how many problems they surface, but by how many they close.

Two tests a beautiful rendering does not pass

This is why we build SpatialWare, our digital-twin operating system for cities, around two questions. First, is the twin computable? Many companies can produce a photorealistic three-dimensional city. The differentiating question is how much analysis can actually be performed using it, from flood hydraulics and infrastructure capacity to cross-domain impact and conflict detection. A twin you can only look at is a map with better lighting.

Second, is the twin alive? A model refreshed once a year is a monument to the city as it was. Keeping a twin current across thousands of square kilometres requires production pipelines that update only what has changed and automate tasks that were once done by hand, including scene reconstruction, segmentation, mesh simplification and texture recovery. Where industry practice has typically automated a fifth to a third of this work, we have pushed that figure beyond 90 per cent, not for elegance, but because a twin that cannot keep pace with the city cannot be trusted to predict it.

The obstacle is institutional before it is technical

Prediction requires cross-departmental data, and the barrier to sharing it is far more often organisational than it is technological. This is why architecture matters as much as algorithms. The AI-powered digital governance framework we are architecting is deliberately federated. It connects existing systems rather than replacing them, synchronises datasets that matter through an AI-enabled orchestration layer, and leaves each department in control of its own data. A city does not need to be rebuilt to become predictive. It needs to be connected in a way that its institutions can work with.

Built at home, because accuracy depends on it

It also needs to be built in the region. A living model of the way a society moves, consumes and behaves is among the most sensitive assets a government can hold. In the Gulf region, where the UAE and Saudi Arabia are investing at national scale in AI strategies and sovereign digital infrastructure, the case for domain-engineered intelligence that is built and governed locally is not protectionism. A generic model trained elsewhere has never seen what a sandstorm can do to a coastal highway or how a city’s evening traffic patterns shift during Ramadan. In this context, sovereignty is not just a matter of control, but a precondition for accuracy.

What will test the loop next

Two shifts will demonstrate whether the sense–understand–decide–act loop holds under pressure. The first is the low-altitude economy: thousands of drones making tens of thousands of trips a day above a city cannot be regulated using a two-dimensional map. A platform that already has a live, three-dimensional picture of the city is the natural place to manage that airspace. The second is agentic AI, which is quietly redefining what a dashboard can be. The next city operations centre will have fewer screens because a copilot that recommends an action and, with human approval, executes it will make most of today’s screens redundant.

The same principle governs the other end of our work. Domia, our AI-native home, learns a household’s rhythm, distinguishes family members by voice and senses presence, posture and even a fall without relying on a single indoor camera. Whether the environment is a villa or an entire emirate, the design goal is the same: a system that anticipates rather than records, and protects without watching.

A new benchmark for public service

For a century, cities have been judged by response time: how quickly a crew arrives or how fast a road reopens. The next benchmark is subtler and more demanding, measured by how many problems residents never experience because a system can see them coming and act before they reach the street. Cities can now predict. The ones that lead will be those that learn to close the full loop.