September 24, 2026
by Arnie Gess, president of Construction Links Network
Artificial intelligence is already becoming familiar on construction projects. Teams are using AI to search documents, summarize information, analyze data, generate reports and help answer questions buried inside increasingly large collections of project information.
But another shift is beginning.
Instead of simply answering a question or generating a response, emerging AI systems are being designed to take actions, coordinate multiple steps and work toward defined objectives within construction workflows.
This is commonly described as agentic AI.
The distinction could become important for construction companies. The next generation of AI may not simply help project teams find information faster. It may increasingly help move work forward.
Most people have experienced generative AI through a chatbot. A user asks a question, the system produces a response and the interaction largely ends there.
AI agents are designed to go further.
An agent can potentially receive an objective, gather relevant information, reason through a sequence of steps, interact with software or data sources and take permitted actions toward completing the task.
In construction, that could mean the difference between asking an AI tool to summarize an RFI and having an agent identify a potential issue, gather the relevant drawings and specifications, prepare an RFI draft, route it for review and monitor what happens next.
The important word here is permitted.
Agentic AI does not necessarily mean giving software unlimited autonomy. The systems now emerging in construction can operate within defined permissions and workflows, with people retaining oversight and approval authority.
The progression can therefore be thought of as moving from chatbot, to AI assistant, to AI agent, and eventually toward connected agentic workflows.
This transition is already becoming visible in construction technology.
Procore has introduced AI agents designed for workflows including RFIs, submittals, daily logs, contract review and project search. Its technology can work across drawings, specifications, photographs and other project information, while actions can be governed by existing security permissions and human approval.
The significance isn’t simply that AI can read construction documents.
It is that software can begin using the information inside those documents to help initiate and coordinate work.
Consider an RFI.
A conventional AI assistant might locate the relevant specification or summarize previous correspondence.
An agent could potentially gather the relevant project information, identify conflicts, prepare an RFI, connect supporting documentation and send the draft to the appropriate person for review.
The project professional still makes the critical decision, but much of the administrative work surrounding that decision could potentially be automated.
That changes the role of AI from information retrieval toward workflow participation.
Imagine a superintendent discovers a field condition that conflicts with the current design.
Traditionally, resolving the issue could require reviewing drawings and specifications, taking photographs, contacting the project team, preparing an RFI, determining whether materials are available, evaluating schedule implications and eventually assessing possible cost impacts.
Those activities may involve several people working across multiple systems.
In an agentic workflow, AI could potentially connect portions of that process.
An agent could analyze the field information against current project documents, retrieve the relevant drawing and specification, prepare an RFI draft and identify the affected schedule activity. Other connected systems or agents could potentially check procurement information, material availability and possible cost implications.
The result could then be presented to the appropriate professionals for review and authorization.
McKinsey recently described a similar construction scenario in which AI agents connect field information with models, drawings, procurement records and schedules before presenting project leaders with possible options and impacts. Human experts remain responsible for approving the solution.
That distinction matters.
The near-term opportunity may not be autonomous construction management. It may be automated coordination surrounding decisions that still require professional judgment.
Construction projects generate enormous amounts of information.
Drawings, specifications, BIM models, schedules, contracts, RFIs, submittals, change orders, photographs, inspection records, safety documentation, emails and field reports can all contain information needed to make a single project decision.
Yet those records frequently live in different systems and are managed by different organizations.
That fragmentation creates administrative work.
Project teams spend significant time finding information, checking which version is current, transferring data between systems, preparing documentation and following up with other participants.
AI agents could be particularly useful where workflows involve predictable sequences of information gathering, analysis, documentation and routing.
Potential applications include estimating and bid preparation, design and specification review, scheduling, procurement, contract administration, document control, progress reporting and project closeout.
McKinsey’s 2026 analysis identified more than 150 workflows across 25 AEC-related domains with varying degrees of AI and automation potential. It argues that near-term opportunities are concentrated in repeatable knowledge and coordination workflows where fragmented information creates delays and inconsistent decisions.
For an industry trying to deliver increasingly complex projects with limited experienced personnel, reducing administrative friction could become increasingly valuable.
There is a catch.
An AI agent can only make useful decisions from information it can reliably access and understand.
Construction data is often fragmented across software platforms, spreadsheets, emails, documents and individual project teams. Information may also be incomplete, duplicated, incorrectly classified or outdated.
Giving AI more autonomy does not solve those problems.
It can make them more important.
An agent working from an obsolete drawing, inaccurate schedule or incomplete procurement record could potentially act on bad information faster than a person would.
That makes connected and well-governed project data an important foundation for agentic AI.
Accenture’s newly launched Construct business illustrates the direction. Its approach to AI-enabled capital-project delivery centres on a common project-data foundation intended to create a shared operational view across project participants and support AI-enabled workflows.
Procore is making a similar connection between its common data environment and AI agents, positioning trusted project information as the foundation agents require to automate workflows effectively.
Before construction companies ask what AI agents can automate, they may increasingly need to ask another question:
Is our project data ready for them?
Construction is not a consequence-free environment.
Decisions can affect worker safety, structural performance, contractual obligations, project cost and schedule.
That makes the boundary between automation and professional judgment particularly important.
AI might prepare a change request, but who authorizes the financial commitment?
It might identify a design conflict, but who approves the engineering solution?
It might flag a safety concern, but who determines whether work can proceed?
It might recommend a schedule change, but who evaluates the contractual and operational consequences?
These are governance questions as much as technology questions.
Procore says its current agentic approach uses human-in-the-loop oversight, with agents operating under permissions and requiring human consent before actions are finalized. McKinsey similarly argues that AEC companies will need clear audit trails, data controls and defined points where people review or approve AI-supported work.
The most useful construction agents may therefore not be those given the most authority. They may be the systems that understand when human authority is required.
Despite the rapid pace of development, construction should be careful not to confuse product announcements with widespread industry adoption.
Many of the capabilities now being introduced are new, expanding or still being developed.
Autodesk, for example, previewed its next generation of agentic AI in September 2026. The company describes a direction in which Autodesk Assistant can use connected project context and coordinate information and capabilities across products, projects and teams rather than operating only within a single application.
Real-world implementations are also beginning to appear. Design-build firm Haskell recently tested Procore’s agentic AI on its headquarters project, with Procore reporting that AI-assisted submittal review reduced review time by approximately an hour in the tested workflow. That is one company example rather than evidence of industry-wide productivity gains, but it demonstrates that agentic capabilities are beginning to move from demonstrations into actual project environments.
The transition will likely be uneven.
Some organizations have connected project data and mature digital workflows. Others still rely heavily on spreadsheets, email and disconnected systems.
Agentic AI will not erase those differences overnight.
For decades, construction software has largely been something people operate.
Someone opens the program, enters information, searches for a document, updates the schedule or generates a report.
Agentic AI introduces a different possibility.
Software could increasingly participate in the workflow itself—monitoring information, responding to events, preparing actions, coordinating tasks across systems and bringing exceptions or important decisions to people when judgment is required.
That does not mean construction companies are about to become autonomous.
Projects remain physical, unpredictable and dependent on people with experience, trade knowledge and professional responsibility.
But the relationship between people and construction software may be changing.
The question may gradually shift from “What can this software tell me?” to “What work can this software help complete?”
The next generation of construction software may not simply tell project teams what they need to know.
It may begin helping them get the work done.
