Most articles about AI in construction read like they were written by someone who has never been on a job site. They promise digital twins, autonomous bulldozers, and end-to-end transformation. Then you go to work the next day and your problem is still that the foreman needs a site safety checklist by 4pm and the file is on someone's WhatsApp.
This post is for the people in that gap. Here's what I've found AI is actually useful for in contracting work right now — boring, practical, time-saving things — and what it still isn't ready for.
What AI is good at today
Drafting and rewriting
Letters to clients, scope clarifications, RFI replies, meeting minutes, follow-up emails. AI is excellent at producing a serviceable first draft that you then edit. The trick is to give it the context — your project, your tone, the previous correspondence — and a clear goal.
This alone will save a project manager an hour a day if used well. It doesn't replace good writing; it replaces the painful blank page.
Summarising long documents
Tender documents, specifications, contracts, meeting transcripts. Anything more than ten pages is a candidate. Modern AI tools can produce a faithful one-page summary of a hundred-page document in seconds.
A caveat: never act on a summary without checking the source. AI summaries are confident-sounding, and when they get something wrong, they get it wrong cleanly enough that you won't notice unless you check.
Extracting structured information
Pulling all the deadlines from a contract. Listing every material referenced in a specification. Identifying every clause that mentions liquidated damages. AI is fast and accurate at this, and it turns documents that were previously read once and filed forever into searchable data.
Translation and tone shifts
Arabic to English and back. Formal to casual. Technical to client-friendly. The current generation of language models handles regional Arabic dialects far better than they used to, though they're still better at MSA than at, say, Saudi or Egyptian colloquial.
Generating lists, checklists, and scaffolds
Site inspection checklists, quality-control lists, risk registers, RACI tables — anything where you start with a topic and want a starting structure to edit. AI gives you a generic-but-reasonable v1 in seconds. Your expertise turns it into the right v1 for your project.
What AI is mediocre at
Numbers
Cost estimation, quantity takeoffs, BOQ pricing. Today's general-purpose AI is unreliable at arithmetic, surprisingly so. It will confidently produce numbers that are wrong, and the errors are not always obvious.
If you're using AI anywhere near pricing, treat it as a structuring assistant — it can help you organise an estimation template — but keep the actual maths in a spreadsheet you control.
Drawings and technical diagrams
Reading a structural drawing, interpreting a section detail, understanding a node connection. Multimodal models are getting better here, but for anything safety-critical, they aren't there yet. Use them to triage drawings (flag pages that mention a particular item) but not to interpret them.
Anything where being wrong is expensive
The classic example: contractual advice. AI can summarise a clause beautifully and still get the legal effect of that clause subtly wrong. For decisions where being wrong costs money or jobs, AI assists; humans decide.
What AI is bad at
Tacit knowledge
The reason your senior engineer is worth ten times what their CV suggests is that they know things that aren't written down. The way a client signs off changes when their CFO is in the room. The fact that this particular consultant always pushes back hard in week three. AI cannot help with any of this. It only knows what's been written down.
Anything physical
Walking the site. Smelling damp concrete. Noticing that a worker isn't wearing his harness properly. The job is still mostly physical, and AI is not.
A simple way to start
If you've never used AI in your work and want a low-risk first project: pick one document you write often, like a weekly progress report. Ask an AI tool to draft it from your bullet points for the next three weeks. Compare what you got versus what you wrote yourself. Adjust the prompt each time. By week three you'll have a good sense of whether it saves you time.
That's the whole game right now. Start small, measure honestly, expand only where the savings are real.