I run the technical office on a residential project of 50 buildings and 1,100 units in Jeddah. On a normal week that means shop drawings, material submittals, method statements, RFIs and a lot of correspondence on Aconex. Most of this work is not design; it is writing, checking and chasing.

Over the last year I started using AI tools for parts of that work. Not to replace engineering judgment, but to get a first draft on the table faster. Here are the five workflows that actually stuck, and the rule I never break.

1. A first draft for every RFI

An RFI has to be short, clear and point to the exact drawing and clause. I give the tool the drawing number, the clause, what we found on site and what we propose, and ask it to structure the question. Then I check every reference myself.

Write a short RFI to the consultant.
Drawing: [drawing no.], detail [x].
Spec clause: [section/clause].
Site condition: [what we found].
Our proposal: [proposal].
Keep it under 120 words, one clear question at the end.

A routine RFI used to start from a blank page. Now the first draft takes a few minutes, and my time goes into checking the drawing references and the proposal, which is the part that matters to the consultant.

2. Method statement skeletons

I paste the relevant spec section and our company template, and ask for a skeleton: scope, references, sequence, equipment, inspection points, HSE. The site-specific sequence, crew and equipment always come from me and the site team.

Waterproofing and blockwork were the first two. The tool gives me the standard headings in seconds; the hold points, crew and equipment are filled in with the site engineer.

3. Checking submittals against the spec

For a material submittal I give the spec clause and the product datasheet and ask for a compliance table: requirement, what the datasheet says, comply or not, and what is missing. It is a checklist to speed up my review, not the review itself.

The compliance table makes gaps obvious: a missing test certificate, a value that does not match the clause, or a datasheet for a different model. I still open the documents and confirm every line before I comment.

4. Turning long threads into action lists

Long email chains and meeting notes become a table of actions, owner and due date. That table goes into our tracker, and I confirm each item against the original before it does.

5. Formulas for reports

Our submittal and RFI status reports run in Power BI. When I need a measure or a formula, I describe the table and the result I want and get a starting point, then test it on real data.

The submittal and RFI status dashboard is a good example: overdue items, items with the consultant, and items returned with comments, by package.

What I do not use it for

• Approving anything. The signature and the responsibility stay with the engineer.
• Calculations that need a licensed check.
• Uploading confidential client or company documents to public tools. I remove names and numbers first, or use the tools the company approves.

Company and client documents stay inside approved tools. When I use a public tool, I remove project names, numbers and anything identifiable first.

The takeaway

AI did not make me a better engineer. It gave me back time on the writing, so I can spend it on the part that needs an engineer: checking, coordinating and deciding.

If I had to put a number on it, it saves me a few hours every week on writing and formatting. Those hours now go to site coordination and reviewing drawings.