Human Accountability: How Construction Leaders Can Scale AI Ethically

Steve Cangiano is the Chief Product Officer at CMiC, a global leader in construction management software.

getty​By now, many construction firms have tried to incorporate AI into their operations; however, fewer use it daily.

As a marketing leader at a construction ERP software company, I’ve watched a familiar pattern emerge across the industry: A firm loads every corporate document into a database, uses a chatbot to get responses to questions and expects results to follow. While some employees find value in that process, most are not there yet.

The gap between pilot and practice is rarely about technology. It is about how AI is introduced, where it is applied and whether people trust what it produces. Here is what I have learned about closing that gap.

In recent years, the most common mistake I see is companies deploying AI without identifying the problem they’re trying to solve or the goal they’re trying to achieve. General-purpose tools ask users to discover their own use cases, and most busy professionals will not take the time to do so.

Instead, start with dedicated agents built for the tasks people already perform manually. These are often routine job-cost and project-accounting activities that consume hours each week. Some examples include importing a long spreadsheet of customer records, simulating how an accounting batch will affect the general ledger before posting it or previewing the impact of a transaction on a project budget. When an agent performs these tasks reliably, adoption often follows because the value is obvious. Dedicated agents can also help newer employees complete tasks in screens they have never used or cannot remember how to find.

What follows is scalability. Once teams understand what standard agents can do, they can start asking better questions or entering better prompts. For example, they may ask, “I have a problem nobody anticipated. Can I build an agent for that?” Giving teams the tools to customize or create their own agents can turn early familiarity into organization-wide use.

At the highest level, leaders often measure AI returns in larger areas or functions, such as cost control or schedule performance. While those gains matter, the most immediate return comes from something simpler: the speed at which the business can answer its own questions.

By design, the answers users are looking for have always lived in the ERP system. What has been missing is access. For some firms, especially those without large IT or analytics teams, an executive who wants to know which regions deliver the highest margins on a specific project type has to request a custom report, which can take a week or more to prepare.

Query agents can change that. A leader can ask in plain language, “What is our margin on tilt-up concrete jobs in Texas?” They can then get an answer in moments. Field teams benefit in the same way when they need a detail buried in a spec book or drawing.

Meanwhile, task agents that create and update records may deliver a quieter return. Each automated task takes a little less time and effort. Over the first 12 to 24 months, that reduced friction adds up, particularly in project accounting and cost control, because people are more willing to complete work that takes less effort.

Historically, construction has been known for slow technology adoption, but things are changing. Here is how I look at it: Working with AI is a skill, much like learning Excel or another application. Prompting an agent resembles giving instructions to a colleague, and many people are not used to articulating requests that way. The learning curve is short, but it is real.

This is where proper design comes in. Rather than sending users to a separate chat tool, embed the agent directly in the screens where work already happens. When the agent appears in a side panel, users can keep their context and work alongside it.

Another barrier to adoption is trust. From what I’ve seen, most users will let an agent take them about 80% of the way through a process and then want to verify the result themselves. That instinct is natural. Every action an agent takes should link directly to the record it created or changed, so a user can review it in one click before finalizing or posting the update. In time, each verified result can build confidence, which, in turn, can drive adoption.

With seasoned employees heading into retirement, many executives are counting on AI to preserve the knowledge those workers take with them. The way I see it, AI will not replace that experience or take over every construction workflow anytime soon. Instead, it will give employees the chance to elevate their performance.

There is a caveat, though. While professionals may use agents to manage volume, they will still need to apply judgment and creativity where it matters. That makes human accountability essential.

Consider a common scenario: concrete yield on a job. An AI system might flag that output per labor hour has declined over the past two weeks, but an alert is not a decision. A project leader still needs to review the data, confirm the trend and determine the cause. The value here lies in flagging the issue earlier than anyone would have noticed otherwise.

This is also the case in software development. Engineers who use AI to write code remain accountable for the code they submit. Construction teams should hold AI-assisted work to the same standard as fully human work.

For leaders ready to invest in AI, the model is simple. Begin with practical tasks your teams already perform, design for trust from day one and keep people accountable for outcomes.

It is critical to remember that a trust-but-verify approach is not a limitation. Rather, it is the foundation for making AI a part of your day-to-day operations.​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?