Yet supervisors are expected to keep work moving, catch quality issues before they become costly, ensure compliance, protect worker safety and deliver projects on time and on budget.

The traditional model doesn't scale.

We've assumed the solution was expanded supervision: more site visits, check-ins and reporting. But even the most experienced supervisor can only be in one place at a time. The problem we should all try to solve is how to extend the reach of the supervisors we already have.

That is where agentic AI has the potential to fundamentally change how construction projects are managed.

Many organizations initially think remote supervision is the challenge. In my experience, it isn't. The challenge is knowing where our attention is needed. The reality is that most crews don't need intervention at any given moment.

Looking for quick answers on construction and engineering topics?
Try Ask ENR, our new smart AI search tool. Ask ENR →

Supervisors spend too much time searching for signals across disconnected updates, phone calls and paperwork. Deloitte’s 2025 outlook described it as a “reactive project controls model" in which teams are burdened by a constant stream of submittals, change events and schedule updates that require manual intervention.

That makes it difficult to distinguish genuine risk from routine activity. One crew calls repeatedly with minor questions while another encounters a significant issue but never asks for help. One project appears to be progressing normally until a small deviation turns into expensive rework days later.

It’s impossible to manage based on gut alone. Managers need operational visibility.

That's especially true as organizations try to increase spans of control. With experienced supervisors in short supply, simply adding more management layers isn't sustainable. Organizations are finding ways to help one person effectively support more crews by giving them better information about where they can have the greatest impact, in the moment when the outcome can still be changed.

We've seen what this looks like in practice. In one deployment with utility company SGN, remote managers were able to oversee substantially more field crews using our Desktop Supervisor module. Within a few weeks, they had enough capacity that the organization assigned them responsibility for a second depot.

Most construction technology documents what already happened, but agentic AI can respond while conditions are still changing.

Construction software has largely served as a system of record. It has helped organizations standardize workflows, capture documentation and understand what happened. More recently, AI has made it easier to search, summarize and analyze that information.

But projects are not won or lost after the fact. They are won or lost while work is underway, when an emerging issue can still be corrected.

Specialized agents can compare live field activity against multiple sources simultaneously: project plans, company procedures, permit requirements, operational data and external regulations. They surface in situations where a supervisor's judgment is needed most.

One of the biggest surprises we've seen is that the highest-value agents do not perform the most complicated tasks. They monitor for lead signals of deviations.

For one utility contractor, we're developing an agent that reviews field video alongside permit information, company-specific rules and statutory right-of-way work requirements. Rather than asking crews to spend 15 minutes documenting every possible compliance detail, the agent focuses on the handful of conditions most likely to result in fines and flags only issues that require corrective action.

In another deployment, an agent compares what crews say they are doing in the field against the approved plan. If work begins to drift from the plan, supervisors are alerted before it has to be redone.

We saw the importance of this during one deployment where a manager manually caught a crew describing a concrete placement that differed significantly from the approved specifications. Left unchecked, the pour deviation could have resulted in millions of dollars in rework. Going forward, this is exactly the type of comparison agents can perform automatically across thousands of jobs.

Just as important, we've learned that these agents have to know when not to make a judgment. 

If the information needed to verify compliance is incomplete, an agent should not assume everything is fine. It should say, "I can't confirm this." On a jobsite, an honest "I don't know" is far more valuable than a false "all clear."

Construction has never lacked data. It has just not been able to connect that information quickly enough to support better decisions. Yet no agent understands project context, customer relationships or field realities the way an experienced supervisor does.

If agentic AI allows one supervisor to support more crews, the opportunity is deciding what to do with the capacity that creates. Organizations can take on more work, respond faster, improve quality or redeploy experienced people where they create the greatest value. They can even choose to invest some of those gains back into the field, creating a better experience for those doing physically demanding work every day.

Agentic AI allows construction leaders to rethink the operating model itself—and that may prove to be its greatest contribution to the industry.

Shelley Copsey is founder and CEO of FYLD, the AI-powered operating system for high-risk fieldwork. Under her leadership, system has become a trusted partner to global infrastructure, utility, energy and construction organizations, helping frontline teams and managers reduce risk, improve productivity and deliver more predictable outcomes through real-time field intelligence.