Construction has a well-earned reputation for being slow to trust new technology. I say that without judgment; I say it as someone who has spent the last eight years sitting at the exact intersection of digital tools, sustainability, and the built environment, across Europe, India and now, UK.
And now, with AI tools popping up faster than national policy and procurement frameworks can keep up, I want to offer a perspective that is neither hype nor fear. Just an honest account of what I have seen work, what worries me, what sounds promising, and where I think this is going - as a practicing sustainability consultant in UK construction industry.
Why construction has been slow in adapting to digital work - and it's not irrational
If you work in this industry, you already know the answer. Construction is not a single organisation with a single data system. It is a temporary coalition of architects, structural engineers, MEP consultants, contractors, clients, and planning authorities - each with different software, different assumptions, and different definitions of what a "model" even means, constantly evolving at different stages.
While developing a national-scale building simulation platform for Switzerland's residential building stock, I learnt that the data pipeline is almost always harder than the model itself. Getting clean, consistent, trustworthy input data across stakeholders is the real engineering challenge. Lately, I have also been evaluating digital tools across the construction sector - testing against established methodologies, and understanding what makes a tool actually useful at the decision-making level.
So when people ask why construction has not adopted digital tools at the pace of, say, financial services, my answer is: the problem is not the technology. It is fragmentation, inconsistent data standards, and the absence of a shared source of truth. AI does not automatically solve any of that. It inherits whatever mess already exists.
What could change - and why it matters
With the given apprehension this industry has about digital tech, something like AI needs to be adapted slowly in construction. For that, we first need to understand what the current tools are doing well (especially, those which are being deployed at a large scale) - the current digital tools being used in construction today automate the parts of the workflow that used to consume most of a consultant's time: pulling data, normalising units, cross-referencing standards, generating scenario outputs. That automation is where AI, done carefully, will start to matter. They are not AI in any generative sense - None of these tools use AI in the way most people picture when they hear the word. There are no chatbots making decisions. What they do is simplify the workflows.
So the next natural question is - can AI, if deployed, process an Environmental Product Declaration (EPD) faster than a graduate engineer. It obviously can. The question is what that graduate engineer does with the time they get back. They automate: pulling data, normalising units, cross-referencing standards, generating scenario outputs. That is already valuable. But the next step — where AI starts to assist with interpretation, not just retrieval — is where the real shift begins.
The clearest way I can illustrate it is through something that happens constantly in sustainability consultancy. A graduate engineer is asked to assess a product's Global Warming Potential (GWP). Traditionally, they open the EPD, locate the GWP figure, check which life-cycle modules are declared, cross-reference it against the project specification, and flag anything that does not align. That process takes time; not because it is intellectually demanding, but because it is slow, repetitive, and easy to get wrong when you are doing it across thirty products on a tight deadline.
The real value: clearing the noise so humans can think
An AI agent — structured properly, with clear ethical guardrails — can do that extraction and cross-referencing in seconds. What it cannot do, and should not try to do, is make the judgment call: is this EPD actually appropriate for this application? Does the declared scope match the way this product is being used on site? Should we be asking the manufacturer for a third-party verified declaration instead?
Those are human questions. They require contextual understanding, professional accountability, and the kind of critical thinking that comes from experience in the field.
This is the framing I want to push back on when I hear "AI will replace consultants." The more accurate picture is: AI handles the retrieval so that consultants can do the reasoning. The danger is if we forget which is which.
Ethical guardrails are not optional
This is where I want to be direct, because this is where a lot of AI deployment in professional services goes wrong.
The most important one: an AI agent must never invent missing data.
In sustainability consultancy, this is not a philosophical point. It is a professional liability point. If an AI tool fills a gap in an EPD with a plausible-sounding figure because the real data was not available, and that figure ends up in a RICS-aligned carbon assessment submitted to a client, the consequences are real. The tool did not sign off on that report. An experienced professional did.
Every output needs a data-quality signal. Users need to know not just what the AI returned, but how confident it is, what assumptions it made, and what still needs human verification. We built this into our own team guidelines: ask agents for reasoning, not just answers. Keep the consultant in control of every decision.
In practice this means prompting matters. Using a system-level instruction like "never generate a value without citing a source; flag all gaps explicitly" is not just good practice — it is the difference between a useful tool and a liability.
The part nobody talks about enough: AI has an environmental cost too
If you work in sustainability, you cannot in good conscience skip this.
AI infrastructure relies on semiconductors that depend on critical minerals - gallium, germanium, indium, palladium, tantalum — with significant extraction and processing impacts. The energy consumption of large language models, particularly for training, is non-trivial, and the supply chains behind AI chips carry the same embodied carbon questions we apply to any other material in a whole life carbon assessment.
This does not mean we should not use AI. It means we should use it with the same scrutiny we apply to anything else. Run older models when the task does not require the latest. Do not use generative AI for tasks that a simple script or spreadsheet can handle. Be honest about the footprint.
There are open-source tools emerging to help measure this - code-level carbon trackers, emissions calculators for ML projects, energy monitoring APIs. We should be using them, or at least asking our AI vendors for transparency on theirs.
Where I think this is going, and what good looks like
Multi-agent systems are the next real development in this space. Rather than one AI tool doing everything, imagine a structured workflow: one agent reads and parses EPDs, another cross-references against project specifications and standards, another generates client-facing summaries with uncertainty ranges clearly flagged, and a human consultant reviews the output and makes the call.
That architecture - specialists in sequence, with human oversight at the decision points — maps well onto how good sustainability consultancy already works. It is not a disruption to the profession. It is a digital version of what we already do when we run peer reviews and quality checks.
What it requires, though, is that our industry invests in data quality, consistent standards, and the governance frameworks to know when AI outputs are trustworthy and when they are not. RICS, LETI, and UKNZBS are doing important work here. AI tools built on top of those frameworks, with proper validation, can accelerate delivery without compromising rigour.
A final thought
The consistent lesson across all of my experience is that tools are only as good as the thinking behind them.
AI is not going to fix construction's data problem on its own. It is not going to replace the judgment of an experienced engineer who has spent years understanding why a building performs differently from what the model predicted. It is not going to take professional accountability off anyone's desk.
But used carefully, with clear ethical boundaries and honest expectations, it can give practitioners more time to do the work that actually matters - the critical thinking, the difficult conversations with clients, the judgment calls that require a human in the loop.
That is worth getting right. And getting it right means being honest about where the limits are.
Note: AI has been used in the writing of this blog to improve grammar, readability, and language clarity. All thoughts and ideas are original.