Michael Fleischman is the Chief Technology Officer and Co-Founder at OpenSpace

Introduction: a moment of convergence

AI has captured the public’s imagination in ways we haven’t seen before. Whether it’s breakthroughs in large language models or the latest in robotics, artificial intelligence is becoming part of everyday conversations—and not just in tech circles.

But why now?

What we’re seeing is the result of two major trends coming together: spatial computing and generative AI. At OpenSpace, we’ve been building at this intersection for years, and we call it the OpenSpace Spatial AI Engine—the core intelligence that powers many of our most advanced features.

Spatial AI combines spatial computing, which interprets the physical world, with generative AI, which interprets language and context, tuned specifically for the realities of construction. It’s the engine behind OpenSpace’s Visual Intelligence Platform, turning raw jobsite imagery into insight teams can act on.

This post takes you under the hood of Spatial AI: what it is, why it matters, and how it’s shaping the future of construction technology.

Why is artificial intelligence advancing so quickly now?

Earlier waves of AI enthusiasm faded because the technology could only handle narrow, scripted tasks. What holds attention today is a structural change: modern systems work across open-ended problems in language and perception, which is why AI has moved from periodic novelty to practical use across industries.

The term artificial intelligence has been around since the 1950s, and interest has come and gone in waves. Occasionally that interest spiked—often aligned with pop culture moments, like the release of 2001: A Space Odyssey in 1968 and The Terminator in 1986.

Despite those early surges, most of AI’s history was relatively quiet in terms of popular impact. There were impressive milestones—like IBM’s Deep Blue defeating chess grandmaster Garry Kasparov in 1996 or Watson winning at Jeopardy!—but these were isolated demonstrations, not the world of the Jetsons that we were promised.

What’s changed in the last decade is the combination of computing power, data availability, and algorithmic sophistication. AI is now more than a lab curiosity—it’s finally moving out of the world of pure science fiction to becoming a practical tool across industries.

What is spatial computing?

Spatial computing is technology that lets machines perceive, navigate, and reason about the physical environment around them, rather than operating only on text or numbers.

The term dates back to the 1980s, but in recent years it’s taken on new meaning thanks to applications like these:

Application area Example
Virtual and augmented reality Apple Vision Pro
Self-driving vehicles Waymo
Robotics and drones Spot from Boston Dynamics
Air quality prediction PurpleAir sensors

While companies like Apple have rebranded spatial computing around immersive experiences (in fact Apple insists that their VR products be called “Spatial Computing”), the concept is much broader. It’s about enabling machines to navigate, perceive, and reason about the world around them.

At OpenSpace, we saw this trend on the rise and further believed that as sensors (especially cameras) became smaller, cheaper, and more powerful, there would be an opportunity to bring spatial computing to construction. By combining these sensors with techniques like SLAM (Simultaneous Localization and Mapping) and Structure from Motion, we could automate site documentation and make jobsite data more accessible than ever before.

That idea became the foundation for OpenSpace, and to date, our users have captured more than 40 billion square feet of jobsite imagery.

What is generative AI?

Generative AI refers to systems that create new content, such as text, images, audio, or video, by learning patterns from large collections of examples.

This is the technology behind a new class of AI chatbots, with ChatGPT the most well-known example. These are far more capable than earlier chatbots like Siri and Alexa, let alone the generation before them, most infamously Microsoft’s Clippy.

What makes today’s AI models so capable?

The key breakthrough was a machine learning architecture called the transformer, which enabled large-scale models to be trained across many machines in parallel. This allowed systems to process vastly more data than ever before.

The resulting foundation models learned to do something deceptively simple: predict the next word in a sentence. But in doing so, they absorbed not just grammar and facts, but also common-sense knowledge about how the world works.

For instance, they can “understand” that if you drop a bottle of water, it might spill—and that spilled water could be slippery. This kind of contextual awareness was notoriously difficult to teach machines in earlier eras of AI.

While such foundation models are powerful, they’re not enough on their own. To make them useful for specific industries or tasks, they need fine-tuning—a process that involves exposing the model to curated examples, often with human feedback.

This is where generative AI becomes practical. Fine-tuned models can perform specialized tasks like summarizing reports and answering questions—if they’re trained properly and grounded in domain knowledge.

But without that tuning, these models can feel vague or generic. As DPR Construction highlighted in a recent report, general-purpose AI tools often fall short when applied to real-world construction scenarios. They’re not always wrong, but they’re rarely insightful.

What is the OpenSpace Spatial AI Engine?

What sets the engine apart is the third ingredient. Spatial computing and generative AI are available to anyone, but neither performs well on construction work without being grounded in it. The engine’s domain tuning, built on real jobsite imagery and field expertise, is what turns general-purpose AI into something that holds up against the specifics of a live project.

We built it by combining the strengths of each world:

Discipline What it contributes
Spatial computing Understanding physical environments
Generative AI Understanding language and context
Domain-specific tuning Accuracy on real construction tasks

That fusion is what lets the Spatial AI Engine do more than store jobsite imagery. It understands it:  automatically autolocating images in space and time, then connecting them to plans, BIM models and workflows. The result is a structured, queryable history of the jobsite, not just a record for it, so you can find what you need and act on it.

Looking ahead: AI that works for you

The goal of Spatial AI isn’t just to make technology smarter—it’s to make your work easier, faster, and more informed. That’s why we’re building systems at OpenSpace that are tuned specifically for construction, trained on real data, and designed to integrate seamlessly into existing workflows.

This is just the beginning. As the technologies behind spatial computing and generative AI continue to evolve, so will our ability to turn them into practical, impactful tools for the built world.

Frequently asked questions

What is the difference between spatial computing and generative AI?

Spatial computing deals with the physical world, helping machines perceive and reason about real environments and the objects in them. Generative AI deals with language and content, producing text, images, or other media from patterns it has learned. They solve different problems, which is why combining them is powerful: one supplies an understanding of the jobsite, the other an understanding of language and context.

What is a foundation model?

A foundation model is a large AI model trained on a broad range of information so it can be adapted to many different tasks. Rather than being built for one narrow job, it learns general patterns first and is then fine-tuned for specific uses, such as construction, with curated examples and human feedback.

Can AI replace construction professionals?

No, and the way Spatial AI is built makes that clear. The system depends on people for the parts machines handle poorly: capturing the site, supplying field expertise, and verifying what the model produces. Its purpose is to remove the slow, manual documentation work around a project so the people running it can spend their judgment where it actually counts, not to substitute for that judgment.

 

Related resources: