Agentic AI is changing project management. Here’s how we use it.
Most of what gets called “AI” in construction is a chatbot with a hard hat on. You ask it a question, it gives you an answer, and then it forgets you exist.
Agentic AI is different — and it’s why the industry is paying attention.
08- Topic
- Agentic AI
- Reading time
- 8 minutes
- Direction
- Human, always
An AI agent doesn’t wait to be asked. It’s given a job — review this contract, check this programme, compare these tenders — and it works through that job step by step, using tools, reading documents, and flagging what matters. It acts. That’s the shift.
How an agent actually works
Strip the buzzword back and an agent is a loop: it takes a goal, breaks it into steps, uses tools to work each step — opening drawings, querying registers, reading contract clauses — checks its own output against the source documents, and escalates what a human needs to see. Then it starts again, because on a live project the documents changed while it was reading them.
The difference from a chatbot is the difference between asking a passer-by for directions and employing a surveyor. One gives you an answer. The other takes responsibility for a task and keeps working it until it’s done — or tells you precisely why it can’t be.
On our projects, that loop runs continuously across four streams — the same four on our AI page: drawing register reconciliation, clarifications and RFI intelligence, tender and scope cross-referencing, and programme risk scanning. Ingest, cross-check, flag, decide. The agents do the first three. People do the fourth.
What’s actually changing
Project management in construction has always had a volume problem. A mid-sized commercial project generates thousands of pages: drawings, specifications, contracts, programmes, RFIs, variations, meeting minutes, progress claims. No human reads all of it. Not properly. Decisions get made on summaries of summaries, and the gaps between documents are where projects bleed money.
AI agents don’t have that problem. They read everything — every drawing note, every spec clause, every contract condition — and they don’t get tired, skim, or assume someone else checked it. That changes four things in practice.
Document review at full depth. An agent can cross-check architectural drawings against structural and services documentation and surface conflicts before they become variations. Scope gaps that used to be discovered on site, at the contractor’s price, get found on paper.
Programme scrutiny that keeps pace. Contractors’ programmes get analysed against actual progress, contract milestones, and historical performance — every update, not just when someone has a spare afternoon. Slippage patterns show up early, while there’s still time to act.
Tender and claim analysis without the blind spots. Agents compare tender submissions line by line against the scope, flag exclusions and tags that shift risk back to the client, and check progress claims against what’s genuinely complete.
Risk that’s monitored, not just registered. Most project risk registers are written once and reviewed quarterly. An agent watches project correspondence and documentation continuously and raises risks as they emerge, not after they’ve landed.
Where agents sit across a project
The volume problem looks different at each phase, so the agents’ work does too.
Design. As documentation develops, the drawing register is reconciled continuously — every revision logged, superseded drawings caught, and cross-discipline conflicts surfaced while they’re a markup, not a variation. Clarifications get drafted from the documents themselves, with an average four-hour turnaround on RFI-grade questions.
Procurement. Tender submissions are cross-referenced against the full scope, not a summary of it. Exclusions, tags and departures that quietly shift risk to you get flagged with the clause they collide with. By tender interviews, your side of the table knows the submissions better than the people who wrote them.
Delivery. Progress claims are checked against the register of what’s actually complete. Programme updates are interrogated for soft float, out-of-sequence logic and durations that history says are optimistic. Contract notices are tracked against their timeframes — which, as we covered in the Superintendent article, is where missed paperwork becomes money.
Closeout. Defects, warranties, manuals and final claims get reconciled against the contract’s actual requirements — the phase where tired teams traditionally stop reading, and where an agent doesn’t know how to be tired.
A live example: tender to recommendation
A tender closes with three submissions on a commercial fitout. The agents ingest the full tender set — drawings, spec, preliminaries, contract departures — and every submission. They cross-check each tenderer’s inclusions against scope, price the exclusions, map the departures against our standard positions, and test each programme’s logic against the site constraints.
What lands on the project director’s desk isn’t a stack of folders. It’s a reconciliation: where each tenderer is genuinely cheaper, where a low number is an exclusion wearing a disguise, which programme survives contact with reality, and a draft recommendation with every finding linked to its source document. The director challenges it, tests the commercial trade-offs, has the conversations — and signs a recommendation they can defend line by line.
The agents found. The people decided. That’s the model on every stream.
What we’ve built at Invero
We’ve trained AI agents to work as part of our project management delivery — not as a gimmick bolted on, but embedded in how we run projects day to day.
Our agents are trained on how client-side project management actually works: Australian standard contracts, procurement strategy, programme logic, cost control, and the commercial realities of commercial and retail delivery. They read every document on a project — all of it — and they work under the direction of an experienced project manager on every engagement.
That last part matters, so let’s be direct about it.
The AI doesn’t run your project. We do.
Anyone telling you an AI agent can replace a project manager is selling something. Here’s what agents can’t do:
They can’t sit across the table from a contractor and negotiate a variation. They can’t read the room in a PCG meeting. They can’t make a judgement call on whether to push a builder or back them, or weigh a commercial trade-off where the contract says one thing and the relationship needs another. They don’t carry accountability — and accountability is the entire point of client-side project management.
What agents do is remove the excuse for missing things. When your project manager walks into a meeting, they’ve effectively had every document on the project read, cross-referenced, and checked. The judgement is human. The preparation is superhuman.
That’s the model: AI does the heavy lifting, experienced people make the decisions, and the client gets both.
What it does to the economics
Here’s the part of the AI conversation the industry avoids: if the tooling makes the work faster and deeper, the client should see that in the fee — not fund a margin improvement dressed up as innovation.
We publish our fees — 1.5–2.5% of contract value for delivery-phase management, fixed at appointment — and the AI back office is a large part of why we can hold that number while reading everything. The depth that used to be a premium service, billed by the hour and rationed accordingly, is now the baseline inside a standard fee. There is no “AI surcharge” line item, and there shouldn’t be on anyone’s proposal. If a firm’s fee didn’t change when their capability did, ask who the tooling is actually working for.
Six questions to ask any AI-enabled project team
Including us. If a project manager tells you they “use AI,” these six questions separate an embedded capability from a subscription and a press release.
1. Who directs the agents? There should be a named, accountable project manager on your engagement — not “the platform.”
2. What do the agents actually read? “Everything on the project, continuously” is the right answer. A tool someone pastes documents into occasionally is a chatbot, not an agent.
3. How is agent output verified before it reaches you? Findings should be linked to source documents and reviewed by a human before they drive a decision.
4. What happens when the agent is wrong? The answer should be boring: a person catches it, because a person checks. If the answer implies nobody checks, leave.
5. Where does your project data sit, and who else can see it? You’re entitled to a specific answer, in writing.
6. Did the fee change when the capability did? Depth going up while fees hold — or published fees at all — tells you who captured the benefit.
Ask us the same six. That’s rather the point.
What this means for you
If you’re a property owner, developer, or tenant delivering a project, here’s the practical difference:
- Fewer surprises. Scope gaps and document conflicts get found early, when they’re cheap to fix.
- Sharper decisions, faster. Questions that used to take a week of digging get answered in hours, with the source documents to back them.
- Genuine oversight of your contractor. Programmes, claims, and variations get scrutinised at a depth that flat fees have never covered before.
- Nothing unread. Every document on your project has actually been reviewed. Ask most project teams if they can say that.
The industry is heading here regardless. Within a few years, running a project without AI-assisted document review and programme analysis will look like running one without email. The question isn’t whether your project team uses these tools — it’s whether they use them well, and whether a human with real experience is directing them.
The bottom line
Agentic AI is the biggest shift in project delivery since digital documentation. Used properly, it makes good project managers significantly better. Used as a substitute for experience, it’s a liability with a subscription fee.
At Invero, we’ve done the work to use it properly. Our agents are trained, tested, and embedded in how we deliver every engagement — and every one of them answers to a project manager who’s accountable to you. Based in Melbourne, working Australia-wide.
If you want to see what that looks like on your project, get in touch or start with a Project Health Check.
Quick answers
What is agentic AI in construction?
Agentic AI means AI agents that are given a job — review a contract, check a programme, compare tenders — and work through it step by step using tools and project documents, flagging what matters, rather than waiting to be asked questions like a chatbot. On Invero projects, agents run continuously across drawing register reconciliation, RFI intelligence, tender cross-referencing and programme risk scanning.
Will AI replace construction project managers?
No. Agents cannot negotiate a variation, read the room in a PCG meeting, weigh commercial trade-offs, or carry accountability — and accountability is the point of client-side project management. What agents remove is the excuse for missing things: the judgement stays human, the preparation becomes superhuman.
Does AI make project management cheaper?
It should show up in value rather than a surcharge. Invero publishes its fees — 1.5–2.5% of contract value for delivery-phase management, fixed at appointment — and the AI back office is a large part of why that number holds while every document gets read. There is no AI line item, and there should not be one on any proposal.
What should I ask a project team that uses AI?
Six things: who directs the agents; what the agents actually read; how output is verified before it reaches you; what happens when the agent is wrong; where your project data sits and who can see it; and whether the fee changed when the capability did.