Practical AI for construction

In pharma, as in much of the construction sector, complexity is high, expectations are higher, and time is always short. Tools need to work—and work quietly. 

That’s why, when AI began to surface as more than just hype, I paid attention. I wasn’t looking for shortcuts, I was looking for support.

When GPT and other AI models became publicly available, I didn’t jump in blindly. I tested them rigorously. I compared how different models responded to technical prompts, safety-related questions, and compliance scenarios. Some gave vague or evasive answers. Others showed real utility.

What struck me wasn’t that AI could “think”—but that it could handle the sort of repetitive admin we usually just live with. Document reviews. Compliance cross-checks. Reformatting. Sorting through folders of reports. These are the slow, draining parts of project delivery that AI is actually good at.

Small wins, big shifts

The first time it really clicked was during a piling project. I needed to confirm that testing had been completed for every pile location. Normally, that would have taken days of searching through spreadsheets and reports. With AI, I fed in the folder and asked a simple question. I had the answer in minutes.

Then I used AI to process a full year’s worth of site diary entries—converting rough logs into a clean, structured report. It saved hours and gave me something I could actually use in project reviews.

I started testing RAMS reviewers. Then I moved into photo-based inputs. Could it spot missing signage? Could it check if a scaffolding image matched the method described in the RAMS? In many cases, it could.

But the key wasn’t just the AI—it was learning how to talk to it. Prompts matter. The better I got at framing the question, the better the outcome. It’s not plug-and-play. It’s a skill, like writing a good brief or managing a difficult contractor.

AI is here to augment, not replace

The more I’ve used AI in project work, the more I’ve become convinced that this isn’t about replacing people. It’s about returning value to the people who matter most: our clients, and our teams.

I’ve always been an advocate of lean delivery—removing waste, improving flow, focusing on value. From the client side, I now see even more clearly that what clients want isn’t novelty. They want outcomes. Certainty. Safety. Simplicity.

AI helps us get there faster—but it still depends on human oversight, context, and experience.

That’s especially important for new graduates. If you don’t understand construction sequencing, constraints, or risk prioritisation yet, AI won’t give you judgement. It won’t explain why we do things a certain way. It can help you format a method statement—but it can’t yet help you write one safely from scratch. You still have to learn the job. Then—when you know what good looks like—AI can help you deliver it faster, with fewer errors, and more clarity.

Security, IP, and the lessons that nearly hurt

I learned very quickly that not all AI tools are safe to use out of the box.

Many public platforms store your prompts on their servers. Some use that data to train future models. Others don’t give you control over where your data goes—or how long it stays there.

For example:

• U.S.-based models like OpenAI must route all data through American infrastructure, which can be subject to federal oversight.
• Chinese-developed models like DeepSeek operate under laws that permit government access to any data stored or processed on their systems.
• Most platforms keep your inputs by default unless you go into settings and manually turn off training and storage.

For construction professionals dealing with sensitive data—especially in regulated sectors like pharma or defence—that’s a real risk. Even simple RAMS details or asset names can become compliance issues if they’re stored in the wrong jurisdiction.

That’s why I’ve focused on using local models when possible, or at the very least locking down cloud tools with strict settings, clear IP policies, and pre-approved workflows.

What I’ve learned

AI in construction doesn’t need to be loud. It needs to be useful.

It’s not about flashy dashboards or trying to replace engineers. It’s about helping us get through the day with fewer bottlenecks and more clarity—so we can focus on leadership, delivery, and doing the job right.

If you’re thinking about using AI on your project, my advice is simple:

• Start small. Pick a repetitive task.
• Don’t share sensitive data until you understand where it’s going.
• Practice prompts. It’s not about tech—it’s about clear communication.
• Keep the human in charge. AI can help—but it doesn’t have judgement. You do.

Used well, AI doesn’t get in the way. It gets out of the way—so your team can do what they do best: build safely, deliver with certainty, and hand over with confidence.