The AI Agent Battle: When Construction Disputes Go Autonomous

Construction is adversarial. Always has been. Probably always will be.

But we're about to add a new dimension to it — and I'm not sure the industry is ready for the conversation.

The Scenario

A subcontractor is in delay. Maybe it's legitimate — unforeseen ground conditions, late design information, a variation that was never properly instructed. Maybe it's not. Either way, their contract requires them to issue notices. Early warnings under NEC4. Notifications of compensation events. Claims for loss and expense under JCT.

Now imagine they've got an AI agent handling their contractual administration. The agent knows the contract inside out. It knows the notice periods, the required formats, the clause references. It's been fed every piece of project correspondence, every programme update, every site diary. It drafts the notice, cross-references the relevant clauses, attaches the supporting evidence, and issues it — perfectly formatted, precisely on time, legally watertight.

The contractor on the other end receives it. Their agent picks it up. Analyses the claim. Cross-references their own records. Identifies weaknesses in the argument, finds contradictory evidence in the daily logs, spots that the notice was issued two days outside the contractual window under clause 61.3. It drafts a robust rebuttal and fires it back.

The subcontractor's agent receives the rebuttal. Identifies the counterarguments. Finds case law that supports its position. Drafts a response that addresses each point, escalates the claim, and references precedent from the Technology and Construction Court.

And now you've got two AI agents in a legal tennis match, each one trained to win for its principal, neither one capable of reading the room.

The Problem Nobody's Talking About

Here's what concerns me.

Construction disputes don't happen in a vacuum. They happen between people who have to keep working together tomorrow. The QS who's pushing the claim still has to sit in the same site office as the PM who's defending against it. The subcontractor who's in dispute on package 4 is also pricing package 7. Relationships matter. Context matters. The things that aren't written down matter.

I've spent thirty years watching how construction disputes actually play out on site. The best contracts managers I've worked with share a common trait: they know when to push hard and when to pick up the phone. They understand that sometimes a strongly worded notice is exactly what's needed, and sometimes it's the worst possible move. They know that winning the battle can mean losing the war — that being technically correct on a £30,000 compensation event isn't worth torching a relationship that's worth £3 million over the next five years.

AI agents don't know any of that.

They'll optimise for the instruction they've been given: protect our contractual position. And they'll do it brilliantly. Every notice perfectly timed. Every rebuttal forensically argued. Every response escalating the stakes by another degree.

Before anyone's had a conversation, you've got two autonomous systems building a dispute file that would make a solicitor's eyes water — and a relationship that's beyond repair.

The Escalation Problem

Think about how disputes actually escalate in construction. It's rarely one big event. It's a slow accumulation of correspondence that gradually shifts from collaborative to defensive to hostile. Each letter a little sharper than the last. Each response a little more entrenched.

I've seen it dozens of times. A reasonable notification turns into a defensive rebuttal. The rebuttal generates a sharper response. Someone uses the phrase "without prejudice to our other rights and remedies" and suddenly the tone has shifted permanently. By the time anyone suggests a meeting, both sides have generated enough paper to make backing down feel like losing face.

Now accelerate that process. Remove the human pause — the moment where someone reads a letter, puts it down, sleeps on it, and decides to call instead of writing back. Replace it with an agent that responds within minutes, comprehensively, with no emotional intelligence and perfect legal recall.

You don't get dispute resolution. You get dispute acceleration.

Two agents can generate more adversarial correspondence in an afternoon than two contracts managers would produce in six months. And every single piece of it becomes evidence if the matter ever reaches adjudication or court. AI construction dispute resolution sounds like a good idea until you realise the AI isn't resolving anything — it's building a case file.

The Imbalance Issue

There's another dimension to this that worries me.

Not every party in a construction contract has equal access to technology. A Tier 1 contractor with a sophisticated AI agent going up against a small specialist subcontractor who's still managing their admin in Excel and Outlook — that's not a fair fight. The Tier 1's agent will issue notices the subcontractor doesn't fully understand, within timeframes they can't match, citing clauses they haven't read since they signed the contract.

We already have a power imbalance in construction procurement. AI agents could widen it dramatically.

Conversely, a well-tooled small contractor with a sharp AI agent could run rings around a larger organisation that hasn't invested in the technology. The dynamics shift in unpredictable ways. I've seen small firms win disputes against Tier 1s purely because their QS knew the contract better — now imagine that knowledge advantage amplified by AI and available around the clock.

So What's The Answer?

I don't think the answer is to avoid using AI for contractual administration. The efficiency gains are too significant, and the reality is that most contractual notices are issued late, incomplete, or not at all — which harms everyone. AI construction management tools that ensure proper contract administration actually protect both parties. The problem isn't AI in construction claims management — it's AI in construction claims management without guardrails.

But there needs to be a human layer. A checkpoint before anything adversarial goes out the door. Not because the AI got the law wrong — it probably didn't — but because the law isn't the only thing that matters.

Here's what I think needs to happen:

Escalation thresholds. AI agents should handle routine notices and administration autonomously, but the moment correspondence shifts into disputed territory, a human needs to review it before it's issued. The agent drafts it. The human decides whether to send it, soften it, or pick up the phone instead. It's the same principle as a site manager who can authorise day works up to a value but needs the PM's sign-off above it — delegation with limits.

Transparency. If your contractual correspondence is being drafted by an AI agent, the other party should know. Not because it's less valid — it isn't — but because it changes the dynamic. If I know I'm arguing with a machine, I'm going to respond differently than if I know there's a person behind the letter who's made a conscious decision to take this position.

Cooling-off protocols. If two agents detect an escalating pattern of adversarial correspondence, they should flag it to their respective principals before continuing. A simple alert: "This exchange is escalating. Do you want to continue by correspondence or arrange a meeting?" That one intervention could prevent more disputes than any amount of clever drafting.

Equal access. The imbalance problem only gets worse if AI contract administration remains the preserve of large firms with big tech budgets. The more accessible these tools become — particularly for small contractors who currently can't afford dedicated commercial managers — the more level the playing field. That's something we think about constantly when building AI into construction workflows.

The Bigger Picture

This isn't unique to construction. Every industry that uses contracts — legal, finance, procurement, property — is heading towards the same collision. But construction is where it'll hit hardest, because construction is where relationships and contracts are most frequently in tension.

We've spent decades trying to make construction less adversarial. Partnering agreements. NEC's spirit of mutual trust and cooperation. Collaborative procurement frameworks. All of that progress could be unwound in months if we hand contractual disputes to autonomous agents without guardrails.

AI in construction administration is inevitable, and broadly it's a good thing. But the adversarial application needs careful thought, genuine industry discussion, and — most importantly — human judgment at the sharp end.

The smartest use of AI in a construction dispute won't be the agent that wins the argument. It'll be the one that knows when to stop arguing.

---

Steve McKenna MCIOB MCIArb is a Chartered Construction Manager, member of the Chartered Institute of Arbitrators, and founder of Construction AI — a purpose-built, AI-native project management platform for the UK construction industry.