Human in the lead: co-intelligence for lead contractors

Who takes the call when a milestone slips?

Not the software. The client calls the project director, and nobody in this industry would have it any other way. I keep returning to that fact because the conversation about AI in construction has drifted toward what agents will do, and away from who answers for it.

Accenture and the Wharton School published research this year on how humans, AI agents and robots are redefining value. It is cross-industry work, and it does not model construction. One line in it stayed with me. The companies getting a return on AI are the ones that keep humans in the lead, not just in the loop. The report calls the operating model co-intelligence: AI that interprets intent, reasons through options and executes bounded work, with people setting direction and owning the outcome. What follows is my reading of it for lead contractors: the firms that hold the contract and answer for the outcome.

In the loop is not the same as in the lead

Being in the loop means being consulted. Being in the lead means setting direction, defining the limits, challenging the analysis, making the trade-offs and owning the result. The report’s view is that leadership becomes more consequential as the tools improve, not less. Lead contractors understood that long before anyone wrote it down.

A lead contractor is the single point where every line of accountability on a project converges. Design coordination, procurement, subcontractor performance, program, commercial exposure, safety and the client relationship all come back to one contract holder. In most businesses a risk-sensitive decision is a line in a ledger. On site, a bad call on a sequence change or a late RFI response can cost weeks of float and a claim that runs to seven figures. An agent can prepare the notice, flag the clash or propose the re-sequence, and on a well-run job it will. What it cannot do is carry the accountability. The more autonomous the tools become, the more explicit that line has to be.

The capacity is trapped in the project office

At the same time, few industries have more capacity locked up in work that earns no margin. The report finds that management, business strategy, writing and recording, and data analysis together account for most of the productivity opportunity it identifies. That is a description of a project office. Anyone who has run one knows where the hours go: finding information across disconnected systems, reconciling versions of drawings and registers, chasing status from subcontractors, writing up what already happened and preparing for the meeting about it.

It is like a site where every instrument works and every reading is accurate, but each one has to be collected by hand and walked upstairs before anyone can decide. None of that work is where a contractor wins or loses a job. The judgment calls are, and they get made late because the people qualified to make them are busy collecting readings.

One record, three kinds of work, and a line between them

The base layer is the project record: current drawings, the contract and its amendments, the program, the submittal and RFI registers, subcontractor commitments, daily reports and the correspondence that qualifies all of it. An agent reasoning from a superseded revision is worse than no agent, because it works fast. The precondition is a single, verified view of the project record, drawn from the systems the contractor already runs.

On top of that record sit three kinds of agent work, and the line between them is where project leadership exercises control:

1. Answering. Agents retrieve, cross-reference and explain which spec clause governs a substitution, what changed between two revisions, which subcontractors are behind on submittals and by how many days. This is where most of the trapped hours live. The person decides what to ask and what to do with the answer.
2. Watching. Agents monitor for the conditions a good project manager wants to know about early: a lead time that no longer fits the sequence, an RFI aging past the contractual response period, a drawing change that touches a procured package. The agent surfaces the condition, the evidence and the exposure. The person decides whether it matters and who needs to act.
3. Acting. Within limits set by project leadership, agents prepare the response: a draft notice, a proposed re-sequence with its effect on float, a recommended allocation of cost to a change event. This is where value and risk scale together, so the limits are set in advance. The person decides what gets signed and what goes out.

Freed capacity is only growth if someone points it somewhere

The report’s warning is that productivity stalls at efficiency unless leaders deliberately redeploy the capacity it frees. In the enterprise case Accenture modeled, the value from growth was more than three times the value from productivity. For a contractor there are three places to point it. On the current project it goes to margin, because earlier sight of program and cost exposure means change events get notified inside the contractual window and claims get evidenced rather than argued. On the next project it goes to pipeline, because delivery record is the pre-qualification currency of this industry and the contractor that can show a governed, human-led way of running a job with agents inside it is the one trusted with the larger work. Across the business it goes to people, because a project engineer whose day is spent on coordination decisions rather than document retrieval is a more valuable professional and a harder one to lose.

Four decisions before the first agent is switched on

Accenture and Wharton close with the observation that co-intelligence does not deliver value by default. It delivers when leaders make deliberate, coordinated decisions and stop initiatives fragmenting. For a contractor executive, I would reduce that to four:

1. Decide at executive level which project outcomes the freed capacity is pointed at: margin recovery, dispute reduction, program certainty or bid win rate. Written down before any tool is looked at, this is the measure every pilot gets judged against. Without it, a pilot proves the technology works and changes nothing.
2. Set the decision rights for agents before the first one is switched on and make the project director accountable for them. This sounds like a reason to wait. It is the opposite. In practice it is one conversation that produces three lists. What agents may do without asking: retrieve, cross-reference, monitor, flag and draft. What they may prepare but a named person must sign: contractual notices, cost allocations to a change event, anything that commits the business to a date or a sum, anything that touches safety. What they do not touch at all, to begin with: the client relationship and the subcontractor negotiation. Add who is called when an agent flags something outside its remit, and how often the lists are reviewed so the first widens as the record and the track record earn it. A contractor that has done this has a tool that acts with authority. One that has not will have a tool that gets quietly switched off the first time it drafts something nobody asked for.
3. Ground every agent in a single, verified view of the project record, because a person can only lead what they can see. This is the decision most often skipped, because it means connecting the systems already in use before adding anything new.
4. Treat the project engineer, the planner and the commercial manager as the people the whole system exists to back up, not as a cost line it exists to remove. They are the ones who will decide whether any of this gets used.

Accountability is construction’s constant

Accenture and Wharton put it in one line: intelligence may be scalable, but accountability is not. For a lead contractor, that has always been the job, and it is why this industry is better placed than most to get co-intelligence right. The people who answer for the outcome already exist, the decision rights already sit with them, and the project is already the unit everything is organized around. What has been missing is a single view of the project record that lets them see the whole of it in time. How a contractor builds that view is the subject of my next article.

When the next milestone slips, how much warning will your project director have had?

Source

The age of co-intelligence: How humans, AI agents and robots are redefining value, Accenture in partnership with the Wharton School, 2026.