Out of Distribution: Why AI fails at engineering

By

Shauna Hurley

and Theodore Galanos

Last Updated May 20, 2026

By

Shauna Hurley

and Theodore Galanos

If AI is advancing so quickly, why can’t it simply learn from drawings, documents and project data and start doing the work of construction teams? Why do we still need workflows and human expertise for engineering work?

Theodore Galanos, Generative AI Lead at Aurecon, explains it this way:

When it comes to AI and engineering specifically, what we’ve really been expecting to happen hasn’t happened yet.

With AI there’s something called distribution and you’re either in or out. Engineering is not yet in distribution. By that I mean models that don’t just scan and summarise, but actually understand and interact with the core aspects of engineering work.

The models are not ‘in distribution’ because they don’t have a meaningful and multimodal understanding of engineering and construction yet. That limitation becomes obvious as soon as you move from general questions to real project work.

Theodore Galanos

Generative AI Leader

Aurecon

“For example, even if you use the strongest model and ask it something about engineering, and you’re like me, ‘pretending’ to be an engineer, you might look at what comes back and think, ‘Wow, that’s amazing,’" Theodore says. "So as a user who doesn’t know the domain, you’re likely to think it’s correct. But if you’re an actual expert engineer, you see that’s not actually the case.” 

“Expert engineers will look at what models come back with at the moment and find a number of things they would never write or do," he says. "So in this way models ‘kind of’ know engineering, but don’t actually know engineering. And that’s very dangerous.”

“Coding is the only task in the world right now that’s in distribution,” Theodore says. “There was an interesting discussion with OpenAI co-founder Ilya Sutskever that highlighted this problem. The singular focus on coding has been a real blind spot in AI.”

From the beginning, the idea was that if we had a model that was superhuman at coding and maths, it would just generalise beyond the world of coding and into other areas and transform whole industries like engineering, architecture and law.

It was a solid idea, but it turns out that’s not happening yet.

When it comes to engineering, for instance, a calculation is not just a calculation. It belongs to a drawing set, a schedule, a climate zone, a standard, a collection of assumptions, and a downstream chain of decisions that may depend on it being right for this case and not some nearby case.

In that environment, generic competence is not enough. It requires understanding, in both a visual and multimodal way, of why things are not in the right place. 

These kinds of tasks are very simple for humans, but very hard for current models.

Theodore Galanos

Generative AI Leader

Aurecon

“That’s why you need to embed human expertise directly," Theodore says. "Tell the model what not to do, define the boundaries of what it doesn’t know, and make explicit the order of operations, checks, guidelines and logic an expert already follows.”

“It’s tempting to treat engineering as one more expert domain that large models will gradually absorb as they become smarter and better trained," he continues. "There is some truth in that. Engineering work is full of technical language, quantitative reasoning, standards, and procedural knowledge, all of which are at least partially representable in text, math, and code. But that description misses the thing that makes the domain difficult in practice.”

We need task-specific data. Think of a Word document. Fifty people collaborate. Comments everywhere. What do we keep? The final version. All that history of collaboration is gone.

From a model’s point of view, that missing history is everything — understanding both what to do and what not to do, what changed, what failed, and why something was rejected.

Theodore Galanos

Generative AI Leader

Aurecon

Until that kind of context is captured and structured, AI will continue to fall short of true understanding. And that brings the focus back to how AI is actually used in practice.

“I think this concept of ‘humans in the loop’ that we’re hearing more and more often really reveals our backwards thinking about human–AI collaboration,” Theodore says. “It implies humans are passive validators, like quality checkers on an AI assembly line.”

“What we actually want is humans driving the loop. If you design the workflow properly, you embed that expertise directly," Theodore says. "This approach also helps ensure expertise and skills don’t atrophy over time. Instead, people can sharpen their thinking and problem solving through this process.”

Rather than expecting AI to replace engineering expertise, the more practical approach today is to encode that expertise into structured workflows and use AI to apply it consistently.

If engineering and AEC are going to benefit meaningfully from AI systems, it needs people who understand the domains, the artifacts, the review practices, and the economics of real workflows to help define what good looks like.

Before this becomes a field of grand claims about what AI can do, it should become a field of good experiments, good environments, and good judgment about where these systems can actually be trusted to deliver.”

Theodore Galanos

Generative AI Leader

Aurecon

Written by

Shauna Hurley

16 articles

Shauna is never short of questions when it comes to construction, tech and science. A professional writer, researcher and podcast producer, she loves sitting down with industry insiders for in-depth interviews that uncover the latest developments, debates and emerging trends. Having worked with organisations like Microsoft and the European Bank of Reconstruction, Shauna joined Procore to explore the complex issues facing construction and share fresh, research-rich insights that help professionals navigate a rapidly evolving industry.

View profile

Theodore Galanos

Generative AI Leader | Aurecon

Theodore Galanos is Generative AI leader at Aurecon and Chief Science Officer at Infrared City, working at the intersection of AI, design and engineering. His work focuses on the “missing middle” of AI — the workflows, systems and orchestration layers that turn model capability into reliable, real-world outcomes, helping organisations apply AI with greater structure, accountability and impact. He regularly shares his latest research and thinking on his industry blog The Harness.

View profile