Imagine a project meeting where the dashboard is green. The purchase order has been raised, the programme has been updated and yesterday’s minutes have been circulated. Then someone asks whether the revised drawing was actually approved for manufacture. Nobody is sure. The supplier has started anyway.
This is an illustrative situation, not a client case. It exposes a familiar distinction: recording activity is different from establishing that the next action is justified. AI can make the records faster. The adviser’s job is to understand which missing fact can make the records misleading.
Experience is valuable when it changes the questions
My view is that a construction AI adviser needs exposure to both successful delivery and difficult outcomes. Successful work teaches what deserves to become a repeatable method. Trouble reveals the assumptions that method depends on: access, authority, payment, sequencing, current information and the capacity of people to respond.
This does not mean that having suffered a loss automatically makes someone a good consultant. Nor must every member of an advisory team have personally managed every possible crisis. The test is whether the team can examine relevant experience honestly, identify what changed and turn the lesson into a check another person can use.
Ask about the bad job. What did the adviser believe at the time? Which record contradicted that belief? When could the problem first have been recognised? What procedure changed afterwards? An answer that consists only of blaming the client, subcontractor or market tells you little about the system they would build.
A variation is a better test than a chatbot demonstration
Take a proposed change to a services installation. The straightforward AI task is summarising the instruction. The useful commercial task is establishing its context. Which drawing revision applies? Who issued the instruction? What work has already happened? What assumptions support the cost? Which programme dependency needs a planner’s attention?
An experienced adviser should separate preparation from authority. The system might assemble a change record and identify missing evidence. A competent person still reviews the position and approves consequential communications. A contract-specific deadline must come from the applicable documents and qualified review, not from an AI model’s confident recollection.
| Situation | Useful system response |
|---|---|
| Clear instruction and current information | Prepare the review pack and route it to the named approver. |
| Conflicting drawings or unclear authority | Expose the conflict, hold the affected action and request clarification. |
| Approved change completed successfully | Capture the actual outcome and update the estimating or delivery lesson. |
The third row matters. A system that records only problems becomes an archive of anxiety. It should also retain what worked, under what conditions and with which evidence, so the next team can repeat a sound method.
Peace begins before the argument
The current TEMRIK Peace guide uses 27 historical rules as a starting point for original construction principles. Its argument is that clearer commitments, shared evidence and controlled disagreement can reduce avoidable uncertainty. It does not claim every conflict can be prevented.
The associated public construction agent playbooks cover contract review, variations, payment evidence, meeting decisions, early warnings and dispute prevention. They give teams something concrete to inspect and adapt. A Markdown procedure still needs permissions and approval controls implemented in the agent platform; words in a file are not an access-control system.
The companion Failure Intelligence construction guide examines 25 failure patterns. Together, these resources invite a better question than “How much administration can we automate?” Ask which misunderstanding, missed decision or repeated omission the workflow should make harder to overlook.
More capable agents make judgement more important
The WebMCP and digital citizens article explores websites exposing structured actions to agents. In a construction setting, an easier route to requesting information or preparing an order is useful only if the business has decided who may perform that action and under what conditions.
Technical capability cannot resolve an ambiguous instruction by itself. The more readily an agent can act, the more carefully its authority, inputs and escalation path need to be designed.
What to ask before appointing an adviser
Ask for an anonymised example of a successful process and one that failed. Then ask to see the proposed workflow tested with a missing document, a disputed instruction and an unavailable approver. Finally, ask how results will be measured after a competent person reviews the output. Preparation time alone can hide a large correction burden.
The CIOB AI Playbook and use-case template offer a useful starting framework for defining objectives, information, risks and evaluation. For another sector perspective, PADISO’s construction AI advisory overview is commercial guidance to compare, not proof of any provider’s results.
My advisory for construction sits within a wider focus on practical AI, project information and business operations. Start with a real decision, establish what a reliable outcome requires, and define a small engagement around it.
The question is not whether an adviser can describe the future. It is whether they can help your team make a better decision when the facts are incomplete—and recognise a good decision well enough to repeat it.
Disclosure: this site, the linked Daniel Roberts advisory service and the TEMRIK resources are associated with Daniel Roberts. External sources are cited for context, not as endorsements. Scenarios in this article are illustrative, not client results.
General educational commentary. Consequential contractual, financial, engineering and safety decisions require competent, appropriately qualified review.