AI won't win your work. But it's already in the room.

Most AEC firms are still asking whether to use AI. The question they should be asking is where to start, and that's where the judgement lives.

Part of the built environment has decided AI is here. The rest is still waiting to be sure. Both are looking at the same technology.

The design side has moved. The latest NBS Digital Construction Report shows that five years ago, fewer than one in ten construction professionals used AI. Now more than two in five have built it into their daily work. A fivefold jump, in a sector that has never rushed at anything. Nearly three in five now worry about being left behind on digital.

The contracting side is less sure. BESA's latest labour market report asked building services firms what AI would do to their work over the next few years. The most common answer wasn't yes or no. It was "don't know". Ask the same firms about the next decade and it flips: more than half expect AI to reshape their roles. The hesitation isn't about whether it's coming. It's that nobody's sure yet what it'll mean, or when.

Same sector, same technology, two completely different moods.

The firms hanging back aren't wrong to be careful. They're just applying site-level caution to places that don't need it.

It's already in how your buyers judge you

Here's the part that should focus the mind, because it's already happening, not coming.

I've written before about how a prospect checks you out before you know they exist. They read your website, your project pages, your people's LinkedIn, they put your name into Google and weigh up what comes back. What's changed in the last eighteen months is that a growing number now get AI to do that work for them. "Who does this kind of project in the North West?" "Summarise what this firm is known for." "Is this practice still active?" And a plain Google search now puts an AI summary above everything else.

So there's a layer of judgement happening about your firm that you never see, run by a tool that only knows what it can find and read. If what's out there is thin, out of date, or so generic it could be any firm, that's the answer it hands back, and you're off the list before anyone picked up the phone. Being legible to AI is just the newest version of being findable. Same job as always: making sure that when someone looks, what they find matches what you'd want them to know.

The question was never really "should we use AI." Your buyers already are. What's left is whether you understand it well enough to be on the right side of it.

The line: scaffolding, not judgement

I've spent the last two years working out where AI is safe to use and where it isn't. First in theory, then in real work, where getting it wrong had a cost. One run of experiments looked at what AI made of changing testing regulations - same question, different briefs, different levels of access to the source material. The outputs swung from genuinely useful to confidently wrong, and the only thing that moved was what I fed it. None of it went near a client. That was the point: to find the line before it mattered.

Here it is. AI is very good at scaffolding. It's bad at judgement. The firms getting value from it know which is which. The ones getting burned haven't worked that out yet.

Scaffolding is the structure underneath the work. First drafts. Turning a rambling bid debrief into clean notes. Research you then check. Reformatting, summarising, getting from a blank page to something you can work with. Tedious, time-consuming, and not where your expertise earns its keep. Hand it over.

Judgement is the part that's actually you. Knowing what a specifier worries about at two in the morning. Which line in a bid a sceptical client will trust, and which one makes them stop reading. The difference between something that's true and something that's true but useless. AI has none of that. It will confidently produce something that looks like judgement, and it will be subtly and expensively wrong.

I do use it on the thinking, but as a sparring partner, not a substitute. I'll get it to argue against my own position so I don't fall in love with it. Useful. But the deciding stays mine.

And the sector already knows where this line sits. Ask those same building services firms where they actually want AI and they don't say "the engineering." They say the admin: the paperwork, the estimating, the risk management. The scaffolding. Nobody's handing it the call that carries the liability. The instinct is already right. It just needs applying everywhere else.

It amplifies whatever's already there

There's a quieter truth underneath the "get the tools, move fast" noise, and nobody selling AI wants to say it out loud.

AI doesn't fix bad work. It speeds it up.

Point it at a clear brief and sharp positioning, and it'll save you hours. Point it at a muddled sense of who you are and who you're for, and it'll produce confident, polished, professional-looking nonsense - faster than you could have managed by hand, and a lot more of it. A firm that can't say what makes it different will now generate ten things a week that don't say what makes it different. The tool didn't create that problem. It just made more of it, faster.

Strategy first, tool second. AI is an accelerant, and an accelerant is only as good as the direction you're pointed in. Sort that out before you reach for it and it's genuinely powerful. Skip it, and you've bought a faster way to produce the wrong thing.

Where to start, if you're starting

Not all AI use carries the same risk. Putting it anywhere near design liability, compliance, or anything that gets built is a serious decision that deserves serious caution. But a lot of what a firm does every day isn't that. First drafts. Bid and tender admin. Research and synthesis. Keeping past clients warm. The proof materials your experts never have time to make - the project sheets, the capability statements, the case studies that are always three years out of date.

This is reviewable, low-stakes, high-volume work, where you see the output before it counts, can check it, and can bin it if it's wrong. A weak draft gets edited, a bad summary gets caught, and nothing's gone out the door that mattered. That's where you start. Not because the work doesn't matter, but because the mistakes are cheap and quick to spot.

One caveat, because it's the objection I'd raise if I were you. "Low-stakes" doesn't mean "no rules." The back office is full of client-confidential material - tenders, contracts, commercial terms - and pasting any of that into a public AI tool is its own kind of risk. The answer isn't to avoid the tools, it's to be deliberate: which ones, what goes into them, what your people are allowed to paste in. Half a day's work to set a policy, not a reason to wait.

Waiting has a cost. It's just an invisible one.

When a firm holds off - waiting for AI to feel safe, or for someone else to prove it - the cost doesn't arrive as a disaster. Nothing breaks. There's no bad day to point at. That's exactly why it's dangerous. The price gets paid somewhere you're not looking.

Here's where. The firm down the road has worked out that AI can turn an hour of meeting notes into a clean summary in minutes. So their people sit in the room and listen instead of scribbling, and walk out with the actions already written up. A few hours a week, per person, handed back. And those hours don't vanish. They go into the work AI can't do: the client thinking, the judgement, the relationships. Week after week, that compounds into a lead. You don't see it happening. You just notice, eventually, that a competitor seems sharper and faster and more on top of things, and you can't quite work out how.

The move is neither to freeze nor to rush. Panic-buying a stack of AI tools is just the no-strategy mistake in a hurry. The move is to start - properly, on the low-risk work, with the thinking done first - and to start now, because the compounding has already begun for the firms who did.

The waiting feels safe. It isn't. It's a cost you've decided not to look at.

If your firm is weighing up AI and you'd rather start where it's genuinely safe than guess, that's a conversation worth having. clare@clarewatson.uk.