We started getting enquiries from construction and real estate development companies about six months ago. Not about the obvious things — virtual tours, generative renders, that kind of use case has been covered. They were asking about something more unglamorous: can AI help us manage the document chaos that comes with a large construction project?
The answer is yes, and the use cases are more interesting than they sound.
The document problem is real and massive
A large construction project in the UAE generates thousands of documents: drawings, specifications, RFIs (requests for information), submittals, change orders, inspection reports, payment applications, subcontractor agreements, NOCs from municipality, approvals from Trakhees or DDA or whichever authority has jurisdiction. These documents have relationships — an RFI response changes a drawing which changes a specification which affects a subcontractor's scope — and in most projects those relationships live in someone's head or in a spreadsheet that's always six days out of date.
We built a document intelligence agent for a Dubai developer that reads incoming project documents, classifies them, extracts key information (the drawing revision number, the RFI response deadline, the change order value, the approval condition), and maps them to the relevant part of the project scope. It doesn't replace the project manager — the project manager still makes decisions. But instead of spending their morning reading through a stack of PDFs, they get a digest of what came in overnight, what changed, and what requires a decision today.
Schedule monitoring: catching drift before it becomes delay
The second thing they asked about was schedule. Construction schedules are notoriously unreliable not because the original plan was wrong but because nobody catches the small slippages early enough. A subcontractor is two days behind on a task that isn't on the critical path today, but it's connected to something that is, and by the time the project manager notices, the delay has cascaded.
Practical agentic AI insights for UAE and GCC leaders — no spam, just what's actually working in production.
We built a monitoring agent that pulls daily progress data from the site management system, compares it against the baseline schedule, and identifies tasks where actual progress is behind plan by more than a defined threshold. It doesn't just flag the task — it traces the dependency chain and estimates the likely impact on milestone dates if the current rate continues. A project manager looking at this in the morning can make a call before a two-day delay becomes a three-week one.
Subcontractor coordination: the endless email chain problem
The third use case is the one that took us longest to scope but has produced the most visible ROI: subcontractor coordination. A large project might have forty-plus subcontractors. Coordinating their schedules, getting progress updates, chasing payment applications, distributing drawings updates, confirming attendance for site inspections — this is a significant coordination overhead that typically falls on the project's administrative staff.
The coordination agent we built handles routine communication: sending drawing updates, requesting progress confirmations, chasing document submissions, distributing meeting minutes, and escalating non-responses. It doesn't make decisions about subcontractor performance — that stays with the project manager — but it removes the administrative layer from their plate and from the admin team's plate, which turns out to be more than you'd think.
What we've learned about construction-specific AI
Construction is document-heavy, relationship-dependent, and jurisdiction-specific in ways that make generic AI tools frustrating to use. A document that says 'as per the approved drawing ref A-103 rev C' is meaningless without the drawing register that tells you what rev C changed. An RFI response that references a 'Trakhees circular from Q3 2025' needs context about what that circular said. These aren't unsolvable problems, but they require building domain-specific knowledge into the system rather than just pointing a general-purpose LLM at a folder of PDFs. That distinction matters in every sector we work in, but it's especially apparent in construction.