AI will transform the industry – if it can overcome 3 human barriers

“To achieve better productivity as an industry, digital is the way forward – and AI is an integral part of that.”

So said Mark Coates, Bentley Systems’ vice president of infrastructure policy advancement, speaking at a recent roundtable the firm held in partnership with the ICE.

“AI is no longer treated as an isolated topic,” Coates added. He reported seeing a “remarkable” growth in confidence in the technology over the preceding six months, especially among senior leaders.

“The genie is out of the bottle and you are not going to get it back in,” he said.

Still, the technology has yet to reach its transformative potential in civil engineering. Human behavioural factors, rather than technical shortcomings, are holding it back.

Revolutionary power

Roundtable attendees agreed that AI tools can perform complex engineering tasks efficiently, often with little input from trained professionals.

An October 2025 survey of infrastructure professionals found that the most commonly explored uses included:

• automating documentation and other routine tasks;
• optimising design workflows;
• improving cost estimation, forecasting and scheduling; and
• monitoring the condition and performance of assets.

Since then, advances in two types of AI – multimodal and agentic – have expanded the technology’s potential applications.

Multimodal AI can process several types of data and forms of content simultaneously. Agentic AI can work autonomously to achieve complex goals set by its users.

Roundtable attendees highlighted AI’s potential in reducing optimism bias among engineers by producing more realistic cost and time estimates for projects. Some even argued that the technology could help to offset persistent skills shortages in the profession.

But they stressed that adopters must remain liable for the quality of all AI-assisted work, supported by robust processes to check it.

Why scaling up AI has been hard

“As we shift from pilots, scaling up AI remains hard,” noted Ian Lumsden, UK digital services leader at Arup.

That’s largely because the industry’s commercial models and processes such as procurement still “aren’t set up for it”, he explained.

“The technology is accelerating, but its success across the industry will ultimately be determined by trust, governance, skills and leadership.”

1. Obstructive procurement practices

While the industry has experimented widely with AI and achieved impressive results in many pilots, attendees noted that it has been slow to apply the technology to large-scale projects.

One barrier to its wider uptake they discussed is the way in which engineering services are typically procured.

Infrastructure clients can be reluctant to depart from industry norms when purchasing large packages of work. In effect, most clients procure engineering expertise by the hour.

This gives the firms supplying that expertise little incentive to use AI tools to reduce their delivery times. In any case, doing that without the client’s permission could be seen as a breach of trust.

Several attendees observed that only senior leaders can resolve this ongoing disconnect between procurement and delivery.

2. Data protectionism

Attendees cited a lack of willingness among organisations to share their data as another limiting factor.

An AI tool is only as good as the data it works with, so will struggle to reach its full potential if its developer cannot obtain relevant high-quality material owned by other organisations. These third parties are unlikely to share it unless they see a genuine benefit in doing so.

There is an opportunity to connect “data sets from different organisations, so that AI can draw insights from across the system of systems”, said Anne-Marie Friel, partner for infrastructure at Pinsent Masons. “But that isn’t really happening yet.”

The reason for this, she explained, is that “all the research and innovation funding is going into technical aspects rather than research into the behavioural aspects of data-sharing and the practicalities of how we all collaborate”.

3. A lack of governance guardrails

The third obstacle discussed at the roundtable is the continuing shortage of standards and processes indicating which tasks can be safely delegated to AI and which ones cannot.

Industry leaders should cooperate in creating a “disciplined delivery framework”, suggested Dr Neda Naghshbandi, technical director for AI and robotics strategy, solutions and delivery at AtkinsRéalis.

Discussing how the benefits and risks of using AI for various tasks can be quantified, she said: “We need to move to an end-to-end process of data flows, workflows, governance and assurance, with all of this being designed to generate real operational value.”

Naghshbandi added: “The whole thrust of this is not about the technology; it’s about the system around that technology and how we are connecting people and processes.”