A new concept is moving through the AI world: intent engineering. It’s the practice of telling an AI system not what to do, but what to accomplish. Instead of writing instructions, you declare a goal. The system figures out the path.

It’s a real advance. But the conversation is happening almost entirely in software circles – about coding agents, product specs, development workflows. The physical world barely appears in it.

The physical world is where intent engineering meets real consequences – pipeline control rooms, logistics networks, infrastructure operations, places where the stakes are measured in safety, reliability, and billions of dollars. And it only works when AI understands one thing that software-focused systems almost never consider: where.

Geography is not a feature. It’s the operating condition of every physical business on earth.

Customers exist somewhere. Assets exist somewhere. Risk emerges somewhere. A pipeline that fails in a river valley behaves differently than one that fails in rocky highland terrain. A parts depot that is technically close to a customer may be hours away in traffic. These are not data problems. They are geography problems – and they require AI that understands location, not just information.

THIS IS WHAT MAKES GEOSPATIAL AI DIFFERENT from the enterprise AI most organizations are building today. Most AI systems are powerful tools for processing information. They can summarize documents, classify content, and automate tasks. But they do not naturally understand where. And for organizations operating in the physical world, that blind spot is consequential.

Consider a U.S. pipeline company – one of the larger operators in the country. It runs 140,000 miles of gas and oil pipelines. Enough to cross the country 40 times.

The company spent $1 billion on pipeline maintenance and upgrades in 2024 – $20 million a week.

But did it spend the $1 billion the right way?

The company has decades of information to tackle that question – incident reports, flow records, maintenance records, sensor data, compliance reports; comparative data about geography, topography, soil, weather, climate.

But it was all on paper.

So they digitized it.

Now the decades of information are something completely different. They’ve become a vast database on which you can do predictive analytics. You can compare things that would have been impossible using paper documents – even if you had those documents side by side on a desk.

Where are dramatic weather events changing the vulnerability of pipelines? Which pipelines have fewer problems – and what qualities make them more robust? Which pipelines suffer the most serious failures? What connects those failures?

Is the key more frequent maintenance? Better weather prediction? A different upgrade schedule? Which maintenance actually makes a difference – and does that vary by location, by type of petroleum product, by age and origin of the infrastructure?

There have always been smart, experienced staff whose job was answering those questions. But they relied on a mix of wisdom, data, and instinct.

Agentic geospatial AI allows those same people to bring science to the question – the science of where.

In the best case, they can ask the system how most effectively to maintain each segment of their network and get answers that lead to better maintenance and fewer failures – without adding cost, or perhaps at lower cost.

But this only works if the geographic data is trustworthy. An AI agent reasoning about pipeline vulnerability needs accurate soil classifications, current weather exposure data, precise records of pipeline location and condition. An agent optimizing depot placement needs real road networks, not approximations. The organizations that will get the most from agentic geospatial AI are the ones that have spent years building and maintaining authoritative geographic data. Now they’re connecting it to the AI systems being asked to act on it. AI reasons only as well as the geographic intelligence it’s been given.

That is intent engineering applied to the physical world. You declare the goal – minimize failures, optimize the maintenance budget, protect the most vulnerable segments – and the system works out the approach. Some recommendations will be familiar. Some will be unexpected. All of them are grounded in geography.

THIS IS WHY GEOSPATIAL AI MATTERS now, at this moment.

The enterprise AI stack is being built in real time – around agents, workflows, and open integration standards. The organizations building that infrastructure are focused on what AI knows. Far fewer are asking where AI operates.

After the Francis Scott Key Bridge collapsed in Baltimore in March 2024, agencies used geospatial AI and drone-derived 3D mapping to create a shared, real-time picture of the wreckage – compressing a recovery process that typically takes months into less than a day. In Chattanooga, Tennessee, geospatial AI directed $6 million in federal grants to exactly the neighborhoods where new tree cover would reduce dangerous surface heat for vulnerable residents.

In both cases, the intent was simple to state. The execution required knowing where.

That is the version of intent engineering that will matter most – not in the development pipeline, but in the physical world where consequences are real, geography is fixed, and where you spend your next dollar determines what happens next.