Artificial intelligence has arrived in the built environment without much of an announcement. It is in the modelling software, in the document management platform, in the take-off tool, and in the email client. Most architecture and interior design practices are already using it several times a day, whether anyone has written a policy about it or not. So, the interesting question is no longer whether to use AI. It is what happens to the work once you do. The part of the story that gets left out Read almost anything written about AI and outsourcing this year and you will find the same quiet admission underneath the enthusiasm: none of it runs on its own. Somebody must prepare the information the system learns from. Somebody must look at what comes out the other end and decide whether it is right. The technology is genuinely impressive. It is also unaccountable on its own. That matters more in our sector than in most. A chatbot that answers a customer badly is an irritation. A model that produces a clash-free-looking coordination report on a badly set up federated model, or a take-off that quietly omits a whole material line, is a cost that lands on site months later, when it is expensive to fix and difficult to argue about. AI is very good at producing work that looks finished. Judging whether it is finished is a different skill, and it is not a technical one. It comes from having drawn the detail before, having priced the package before, having been on the receiving end of the RFI when it went wrong. What this looks like on a real project A Revit technician who understands how a model is meant to be structured can use automation to strip hours out of repetitive family creation, sheet setup and annotation, and spend the recovered time on the parts that need thought. A technician who does not understand the model will use the same tools to produce a great deal of tidy-looking rubbish, very quickly. An estimator with fifteen years behind them can run an automated material take-off, scan the output and know within a minute that the ceiling quantities are wrong because the tool has read a plenum as a room. The software will not flag that. Nothing in the output will look unusual. The estimator knows because they have seen a plenum before. The same holds for shop drawings, submittal review, document registers and coordination. In each case, the tool moves faster than a person could. In each case, someone must own the answer. What to actually look for If AI raises the value of judgement rather than lowering it, then the way firms buy resource ought to change with it. A few things are worth asking of anyone providing you with people: Can they scale up and down at the pace your workload genuinely moves, rather than at the pace of a twelve-month commitment? Do you know what you are paying and what you are getting for it, without unpicking a rate card? Do the people they provide work inside your systems, your file structure and your standards, or alongside them? And when something is wrong, is there a named person who is accountable for it? None of those questions mention AI. That is rather the point. The tools will keep changing. What protects the project is the quality of the person using them and the clarity of who answers for the output. Where ADDMORE sits We are not an AI company in the way that phrase is usually meant. We did not build a model and then go looking for somewhere to point it. We started from the work of documenting buildings, pricing packages and keeping projects moving, and everything we have built since has come out of that. What we provide is people, and the technology those people need to do the work properly. That means qualified back office professionals: 3D modellers, Revit and AutoCAD technicians, US estimators and document controllers, working as an extension of your own office, on your systems and to your standards. It also means business technology support: data engineers, analysts, BI specialists and AI engineers who build and maintain the automated parts of the workflow, the take-off routines, the document pipelines, the reporting behind the reporting. Done well, you never have to think about any of it. Done badly, it is the reason a number was wrong three months ago and nobody noticed. We also build software of our own, including PromoreIQ, our AI platform. We would rather tell you what a tool can and cannot do having had to make one work ourselves. Our people use AI in their daily work, and we expect them to. What we are providing, across all of it, is the experience that sits behind the output: the person who knows what the drawing is for, what the client will ask next, and what a plausible-looking result should have said instead. That is also why we have not rushed to relabel what we do. The work has not changed as much as the vocabulary around it has. Buildings still have to be documented properly. Packages still have to be priced accurately. Somebody still has to be answerable when they are not. The short version AI has made the first draft of almost everything affordable. It has not made the judgement inexpensive, and in the built environment, the judgement is where the risk lives. The firms that get the most out of these tools over the next few years will not be the ones with the largest technology budget. They will be the ones with enough experienced people to tell good output from bad, fast enough for it to matter. If you are working out how to add that capacity without adding overhead, we would be glad to talk it through.