Moving from AI promise to real-world infrastructure reality…
The public discourse on AI would have us believe that we are on the verge of ending the need for human involvement across many fields of technical expertise. In the built environment, that narrative can sound particularly seductive – AI as a way to transform how we design buildings, manage ageing utilities infrastructure, operate transport networks and improve customer experience across complex public systems.
But our experience across infrastructure and asset-intensive environments points to a more grounded reality. Working across disciplines, geographies and asset types shows that these sectors are characterised by sparse or incomplete data, hidden complexity, competing requirements, political turbulence and constrained supply chains. These are not challenges that can be taken away through the wave of an AI wand.
That is why the most useful role for AI in infrastructure is not as a replacement for professional judgement, but as a way to help people make better-informed decisions in complex, high-consequence environments.
Peel away the layers and most projects within the physical asset world boil down to two main components – management of risk and understanding what represents an acceptable risk level.
The sweet spot: low cognitive, high volume
We will start with where AI is delivering genuine value. We tend to describe use cases in tiers, and the tier delivering practical value today is what we would call low cognitive, high volume: tasks where success depends not on an accumulation of engineering expertise but on grinding through a repetitive process reliably and at scale. Classifying thousands of asset records. Checking a design document against a set of standards. Extracting structured data from unstructured reports. Checking drawings for consistency. These tasks rarely make headlines, but they consume enormous quantities of professional time, and they are often a dull and unfulfilling part of the job for the people who do them.
That last point matters more than it first appears, because it reframes the anxiety that dominates so much AI commentary: the fear of job losses, particularly for early-career professionals. Much of this low-cognitive, high-volume work has traditionally fallen to people in junior roles. Yet those same tasks are usually the ones that offer the least genuine learning and growth. Manually reconciling the same kind of dataset for the hundredth time builds very little engineering judgement.
When AI takes on that grind, it does not remove young engineers; it frees them. The time that would have gone into repetitive processing can be reinvested into the work that actually builds expertise: exposure to real projects, to clients, to complex decisions, to technical leadership.
Handled well, AI can help early-career professionals move sooner into higher-value work. In that sense it is best understood as an accelerator, not a replacement and organisations that understand this becomes more attractive places for young professionals to build their careers.
It was never a zero-sum game
There is a deeper assumption buried in the “AI will take our jobs” narrative that this is a zero-sum game. If a machine does a task that a person used to do, the person must therefore do less (and hence we need fewer people). In some industries that logic may carry some truth. In infrastructure it largely collapses, because our world is chronically, structurally resource-constrained. We are not sitting on a comfortable surplus of engineers with time on their hands; we have far more work than we can currently deliver.
The scale of that constraint is worth stating plainly. McKinsey estimates the world needs a cumulative $106 trillion of infrastructure investment through 20401 to keep pace with population growth, the energy transition and technological change. This spans everything from roads, bridges and power grids to the data centres and networks of the digital age. Set against an engineering and technical workforce that cannot grow anywhere near fast enough, that is not fundamentally a demand problem. It is a delivery-capacity problem. This changes the meaning of efficiency entirely. When AI releases capacity in a resource-constrained system, that capacity does not evaporate into idle time or redundancy. It is converted into more delivered scope, more of the backlog tackled, more assets assessed, more resilience designed in. The efficiency is released, but it is immediately reinvested in doing more of the work the world urgently needs, rather than doing the same work with fewer people.
To put this into context, we recently worked with a railway operator to embed AI into their Maintenance Management system to reduce the amount of time their engineers needed to spend reviewing work recommendations. The outcome is not a reduced need for engineering input, but a better use of it. That time freed up for the reviewing engineers is now invested more into a deeper consideration of the best scoping and packaging of works to achieve the best outcomes for the asset. It is a transfer of human time to higher-value tasks.
Risk, judgement and accountability stay human
Which brings us to the heart of it, and the reason we are confident that infrastructure AI will remain human-led. When it comes to large infrastructure, most decisions ultimately come down to the management and acceptability of risk. And risk, in this domain, is a fundamentally human matter.
It is human for reasons of empathy and understanding – judging what level of risk is acceptable to the people who will live beside a structure, travel across it, or depend on it for clean water is an act of judgement rooted in context, values and lived consequence, not a calculation. And it is human for reasons of accountability: when a decision carries the potential for real-world harm, someone must own it.
Accountability cannot be delegated to a model. A professional puts their name, their reputation and their duty of care behind a decision in a way an algorithm never can.
This is why the most impactful applications of AI are and will be people-first: AI must amplify expertise, never replace it. The most valuable role for AI in high-consequence engineering is as a support for human expertise, not a substitute for it. AI can be valuable for surfacing options, running the numbers, testing scenarios, flagging what a human might miss, and then handing a better-informed decision to a professional who remains firmly accountable for it. That means AI initiatives should be judged not only by whether they are technically effective, but by their wider consequences: do they support the people using them, improve the quality and reach of human expertise, and help deliver better outcomes for the communities and environments affected by infrastructure decisions?
What this looks like in practice
None of this is theoretical; the same pattern shows up across real-world applications.
Asset optimisation: In asset investment planning for rail and water networks, the value lies not a machine deciding what to renew or replace. It is AI processing vast volumes of condition and cost data so that engineers can consider a broader range of options and help inform investment decisions. The judgement stays with the engineer; the reach of that judgement can be extended.
Scenario and resilience modelling: For example, with work on climate-resilient transport, or hydraulic and surge modelling for water systems, AI can let us explore far more futures than we ever could by hand: more climate scenarios, more failure modes, more “what ifs”. But choosing which scenarios matter, and what to do about them, remains a human act of interpretation.
Digital twins: Living, data-fed models of physical assets provide a richer, more current picture that helps human operators and engineers make more informed and efficient decisions across a larger portfolio than they could otherwise effectively oversee. In every case the pattern is the same: the technology extends human reach; it does not remove the human.
AI as a catalyst
So how should we think about AI in infrastructure, once we set the hype aside? Not as a replacement for engineers, and not as a threat to the next generation entering the profession.
The most useful framing we have found is AI as a catalyst, something that accelerates a reaction without being consumed by it. It speeds up the work, frees up expertise and allows human judgement to be applied across a far greater volume than is possible today. Pointed at a backlog created by decades of under-investment and now compounded by the demands of climate resilience and evolving needs, that is no small thing.
The firms and public bodies that get the most from AI over the next decade will not necessarily be those chasing the most advanced models. They will be the ones who invest in their data, integrate AI honestly into how work really happens, and keep human expertise and accountability firmly at the centre.
In the real world, AI in infrastructure will remain human-led and data-driven, and in our experience, that is precisely what makes it worth pursuing.
