Competence, clients and the first ninety days
The closing part: how to train professionals without hollowing them out, what to tell clients and insurers about AI use, and a ninety day plan.
Blog
Notes on AI in construction
Practical, no-hype writing on automation, site reporting, data and adoption.
Stop measuring AI by time saved
Hours saved is an input, not an outcome. Six categories that measure what changed in the business, and a six week cycle that produces evidence.
Build an AI that watches work, never one that watches people
One assistant watching a whole project produces noise nobody acts on. Narrow discipline agents, a supervisory layer, and the line that must not be crossed.
The most valuable AI on a project may save nobody any time
Commercial management is an evidence game. A challenge function that adds twenty minutes and moves a reported position ten points beats hours saved.
Design management is a dependency problem, not a document problem
The costly design failures are not one discipline being wrong. They are two disciplines each assuming something about the other. Where AI helps.
From Minutes Saved to Value Realised: What Automated Reporting Actually Saves
Controlled studies put the measured effect of AI between minus 19 and plus 140 per cent. Why a task saving shrinks across a job, and what to ask a supplier.
The project manager who compressed the work, not the thinking
A Wednesday morning with a revised programme, an ambiguous specification and a payment application. What to ask for, what not to, and where judgement stays.
Construction AI does not usually fail by inventing things
Hallucination gets the attention. Real project failures are failures of context: superseded revisions, missed amendments and asserted causation.
What meaningful human involvement now has to mean
Since February 2026 a decision is solely automated if no meaningful human involvement exists. A seven field matrix that makes oversight real.
Classify by consequence, not by which logo is on the screen
Enterprise tools produce wrong answers too. How to classify AI use by the consequence of being wrong and how late the error surfaces, with worked examples.
Build the register before you write the policy
A policy written without knowing what is happening governs an imaginary firm. How to run a four week AI discovery exercise and build a register people trust.
AI can read the contract. It cannot become the dutyholder
UK construction is built around named people carrying statutory duties. That architecture, not human exceptionalism, is why the professions survive.