Standing outside the office of a global leader in Construction Management, I watched a troubled-looking project manager walk out of a meeting, grabbing his phone and immediately summoning some poor soul on his team to get every insight and data point they could on a project that had clearly been hanging by a thread. I knew something already, having already been in several meetings with this company, that he had not one chance of getting that information, certainly not quickly, and the accuracy I'm not sure even the data he pulled out could be questioned as credible or not by the people asking for it. In an alternate world, maybe not one too far away, there's a world where no panic, no frantic calls—just a smooth adjustment to the day's work, orchestrated by...
Breaking the Reactive Mould
Traditional AI responds when prodded. Ask a question, get an answer. Upload data, receive analysis. It's fundamentally reactive.
Agentic AI turns this model on its head. It monitors situations, spots potential issues, and takes steps to address them before they become problems. It's the difference between a calculator and a financial advisor—one waits for inputs, the other keeps an eye on your money and rings you up when something needs attention.
During a recent chat with one of our senior AI advisors, they described this shift as "moving from AI that tells you there's a fire to AI that's already called the brigade and started evacuating the building".
In essence, instead of asking a question on your data, you'll mention a strategy, something you're trying to achieve, a broader goal, the prompt will be dissected and the agentic system will use the right model at the right time to query the correct data and come back to you once it's finished aggregating, refining, optimising, validating and ultimately delivering to the user.
Real-World Applications Taking Hold
The built environment already offers fascinating examples of this technology finding its feet:
At a major UK infrastructure project, BuildPrompt supports clients to automate tracking of requirements against delivery. Enabling compliance monitoring once meant drowning in paperwork. Soon, their agentic system tracks regulatory requirements in real-time, flagging missed inspections before they trigger delays.
A European airport operator I visited has transformed thousands of incident reports into a "living safety brain". Rather than gathering dust in archives, lessons from past incidents actively inform current incidents.
In the US, BuildPrompt is articulating risks across real estate portfolios, helping understand expected and unexpected risk. Delivering insight and ensuring upcoming company strategies are reflected at a granular level in their real estate data.
Technological Convergence Fuelling Growth
Several advances are converging to make these capabilities possible:
The reasoning capacity of AI has leapt forward dramatically. When attending a demonstration at a tech conference in Birmingham last autumn, I watched as a system worked through a complex scheduling problem step-by-step, explaining its thinking at each stage—much like an experienced project planner would.
Multimodal processing means these systems now understand construction drawings, site photos, and written specifications simultaneously. A project manager in Manchester showed me how their system could spot discrepancies between architectural plans and actual site conditions from ordinary smartphone photos.
Hardware improvements have brought powerful models to ordinary devices. "Two years ago, this would have needed a server farm," an engineer told me, pointing to a standard laptop running sophisticated predictive analytics on-site.
Challenges Remain
Construction and infrastructure projects are notorious for running over schedule and budget - typically 20% longer than planned and up to 80% over budget according to McKinsey. Meanwhile, 96% of construction data goes unused, buried in siloed reports and unwieldy documents. Teams waste countless hours searching through emails and spreadsheets for critical information, leading to inefficient workflows and duplicated efforts. The built environment needs a smarter approach to navigate these persistent challenges.
Implementation Challenges:
-
Talent skill gaps (cited by 46% of leaders as a barrier to AI adoption)
-
Cost uncertainty and difficulty predicting ROI for AI at scale
-
Leadership alignment issues (getting consensus on AI strategy)
-
Supply chain dependencies (particularly for hardware/infrastructure)
-
Lack of explainability in AI systems
Employee Readiness and Training:
-
48% of employees rank training as the most important factor for AI adoption
-
Nearly half of employees report receiving minimal to moderate support for AI capability building
-
Millennials (35-44 age group) show the highest AI expertise and can serve as champions
Risk Management:
-
Only 39% of C-suite leaders use benchmarks to evaluate their AI systems
-
Most benchmarking focuses on performance metrics rather than ethical considerations
-
Top employee concerns: cybersecurity (51%), inaccuracy (50%), and data privacy (43%)
From Incremental to Transformative
What's becoming clear from my conversations across the sector is that the most successful organisations aren't treating AI as a bolt-on productivity tool. They're fundamentally reimagining their operations.
A real estate portfolio manager in Edinburgh described how their approach to risk management has completely changed: "We used to have weekly risk meetings where we'd react to problems. Now our system continuously monitors twenty different risk factors and only pulls us in when patterns emerge that need human judgment."
The Path Forward
As I write this from a train heading back down to London, I reflect on dozens of conversations with people at the coalface of this technological shift. The consensus seems clear: agentic AI represents not just a technical advancement but a fundamental reimagining of how we work.
The organisations pulling ahead aren't necessarily those with the biggest technology budgets, but those creating cultures where human expertise and AI capabilities complement each other. They're investing in training, developing clear governance frameworks, and maintaining feedback loops that help systems improve.
The concrete hasn't yet set on how this will transform our industries, but the foundations are clearly visible. And from what I've witnessed on sites across Britain, the building work is well and truly underway.