I wrote this article using AI as a thinking partner. I am disclosing that upfront, because it is relevant to the argument I am about to make.
The construction industry is currently in the middle of a conversation about artificial intelligence — and like most conversations about disruptive technology, it is happening at two extremes. On one side: AI will replace contract managers, delay experts, claims consultants. On the other: AI is a gimmick, irrelevant to the realities of a construction site. Both positions are wrong.
"AI needs inputs. And AI doesn't know what happened on site."
That is the sentence that cuts through both arguments. It is also, I think, the most honest description of where we actually are — and where the real value of human expertise lies in a world where machines can draft notices, summarise contracts and generate delay narratives in seconds.
What AI can actually do
Let me be direct about what AI does well in construction contract management, because I use it regularly and I would be dishonest to pretend otherwise.
AI can draft a notice of delay in the correct FIDIC format in under a minute. It can summarise a 200-page contract and flag potentially onerous clauses. It can explain the difference between a global claim and a particularised claim, outline the SCL Protocol approach to concurrent delay, or generate a first draft of an Extension of Time claim narrative — all faster and more fluently than most human practitioners.
Used well, it is an extraordinary tool. It compresses hours of drafting into minutes. It catches things you might miss when you are tired at the end of a long site day. It is, genuinely, a capable partner for someone who already knows what they are doing.
What AI cannot do
Here is the problem. AI can write the notice. It cannot tell you:
- That the Engineer's instruction last Tuesday constituted a variation — because it knows nothing about Tuesday
- That the access delay in March was actually caused by the Employer's late approval of shop drawings, not by weather — because it was not there in March
- That the subcontractor's claim is inflated because their productivity records don't match what was actually happening on site — because it has never been to your site
- That the relationship with the Engineer is fragile right now and a formal notice might trigger an escalation that serves nobody — because it cannot read the room
- That the contemporaneous records from weeks 12 to 18 are incomplete and need to be reconstructed before any claim is submitted — because it hasn't read your site diaries
These are not peripheral concerns. They are the substance of the work. A notice sent without understanding the underlying event is not just useless — it can actively damage your position.
The input problem
I have spent seventeen years on construction sites across Asia and Europe. In that time, the single most consistent factor separating successful claims from failed ones has not been the quality of the legal argument. It has been the quality of the factual record — and the understanding of what actually happened.
A delay expert with AI can produce a beautifully structured, legally coherent EOT claim in a fraction of the time it used to take. But if the inputs are wrong — if the delay events are misidentified, the programme baseline is incorrect, or the causal link between event and impact is not properly understood — the output is worthless, regardless of how well it is written.
This is not a new problem. It is the same problem that existed before AI. What AI changes is the speed at which a bad claim can be produced. That is not necessarily progress.
The steam engine moment
There is a useful historical parallel here. When the steam engine arrived, many workers responded by smashing the machines — the Luddites, famously, believed that destroying the technology would protect their livelihoods. They were wrong. The people who thrived were not those who fought the machine, but those who understood it well enough to become its masters: the engineers, the operators, the people who could extract from it what it could not extract from itself.
AI in construction is the same moment. The practitioners who will struggle are those at both extremes — those who refuse to engage with the technology at all, and those who surrender their judgment to it entirely. The ones who will thrive are those who understand what AI can and cannot do, and who position themselves accordingly.
The danger of inexperience
There is a risk that concerns me more than AI replacing experienced practitioners. It is the opposite: organisations hiring inexperienced people who lean on AI to compensate for what they do not yet know.
I have seen this play out in practice. A junior team member — not entirely comfortable in English — uses AI to review and respond to a contractual document. The AI produces something grammatically correct and structurally plausible. It gets submitted. The Engineer rejects it. And when I read the Engineer's response, it is obvious — from the structure, the phrasing, the way three different points are addressed identically — that the Engineer's team has run the document through an AI filter too. The AI responded to the AI. Nobody who actually understood the project was in the room.
The document was rejected not because the contractual position was wrong, but because the query was wrong. Because the person reviewing it did not have the experience to understand what was actually being asked — and so the AI, faithful to its instructions, answered a different question. That kind of error does not just waste time. It damages credibility, hardens positions, and can create a dispute where there was none.
The balance I am trying to strike
When I think about how I work — and how I want to work as AI tools develop — I think of it as a balance between two things that are not in opposition but are often treated as if they are.
On one side: the deep, contextual, irreplaceable knowledge that comes from being on site. Knowing what the Engineer said at the progress meeting. Understanding why the subcontractor's crew was pulled off the work in week 23. Recognising that the delay in the foundation works was not just a productivity issue but a direct consequence of the Employer's late submission of the geotechnical report. Reading the room — understanding whether the tone of a meeting signals flexibility or entrenchment, whether a negotiation is genuinely open or already decided. These are things that exist only in the heads of people who were there.
On the other side: the speed, consistency and analytical power that AI brings to drafting, summarising and structuring. Used well, it compresses hours into minutes and catches things you miss when you are tired at the end of a long site day.
The best analogy I have found is motorsport. Put an experienced Formula 1 driver in a sports car and put an amateur in an F1 car. The professional will still be faster — because the skill, the judgment, the ability to read the track and push the machine to its limits, lives in the driver, not the vehicle. AI is the better car. But a better car driven badly is still slower than a good driver in a reasonable one.
This is still a profession where experience matters — where setting the scene for a claim, reading the cultural dynamics of a negotiation, understanding the tone to use with a particular Engineer on a particular day, are the difference between resolution and escalation. AI can help develop the strategy. But you have to know what the plan is before you can ask AI to help you execute it.
"AI can help you develop the strategy. But you need to know what the plan is first."
That is the honest version of where we are. And for what it is worth — I wrote this article with AI as a sounding board. The ideas, the structure, the arguments are mine — we brainstormed them together, challenged each other's positions, and explored the pros and cons. AI helped organise my thoughts into a coherent flow. Then I rewrote that version in my own words, filtered it through my own experience, and gave it a final polish. That, I think, is exactly how it should work.