The Mirage of Competence: How AI Interferes With Your Client Relationships Without Meaning To

This week a designer told me about an enquiry she’d recently received from a prospective client. It was precise, and used language she would expect from someone with real expertise in the field. The client asked for a technical and highly sophisticated solution for their residential project. Switching up a gear, she wrote back with level of detail the brief invited. It was only when the conversation moved from writing to the phone, that she discovered the client was a complete novice, and the brief itself was entirely unsuited to what he actually needed.

I’ve spent the best part of three years arguing that designers ought to be using AI more, not less. I still believe that. I led a CPD webinar on AI for the BIID this week, and noticed that the conversation in the room is changing with time: not so much what AI can do for a designer’s practice, but what it is already doing, unasked, inside the client relationship. AI is interfering in ways new enough that we haven’t yet evolved strategies to handle them. Most of this isn't malicious. None of it is even, strictly speaking, AI behaving badly. It is AI behaving exactly as it always does: dropped into a relationship it has no way of understanding.

Total fluency, zero situational awareness.

This description names the problem. AI writes with the same unwavering confidence whether it knows a great deal about the situation or nothing at all. It cannot tell the difference between a question asked by an expert and a question asked by someone parroting language they don't actually understand, because the words look identical on the page. A client who has read four AI-generated paragraphs on “biophilic design principles” can produce a brief that sounds indistinguishable from one written by a client understands it from long-standing awareness and lived experience. The fluency is real. The situational awareness behind it is not.

This single mechanism shows up at every stage of the client relationship, and in this blog I’ll trace it through, because the response that works at one stage will not necessarily work at another.

At the brief.

The novice-who-sounds-expert problem is the cleanest version of the mirage of competence. We have always used a client’s command of language as a rough proxy for their command of the subject. A client who could talk fluently about junctions and specification was, until recently, reliably a client with some experience of construction or design. That correlation has now broken, and the first time you experience it could be discombobulating, and could lead you to lose a project without you ever really understanding why. A fluent brief can no longer be trusted as a signal of an experienced client.

We have a duty of care to our clients which must now include a deliberate but subtle habit of verifying the understanding behind the language before pitching a response at the level the brief implies.

There is a second manifestation at the briefing stage, and it is even more wicked (in terms of degree of difficulty rather than evil intent) than the first, because it arrives as a fact. This is the client who comes to the table already wedded to an AI-generated concept for their home, one that defies their true needs, building regulations, the physical dimensions of the room it is meant to occupy, or, on occasion, gravity. The novice-client-who-sounds-expert problem at least leaves room for a conversation, because the client knows, somewhere, that they are behind the curve. This version does not. The client has already seen the render. It is beautiful, it is specific, and it is theirs in a way a verbal brief never quite is, which means the task is no longer to establish what they want, but to dismantle, gently, something they have already fallen in love with.

Pointing out that AI was never given the room’s dimensions, the load-bearing wall, or the local planning history lands very differently depending on whether it sounds like a correction or a concession. It needs to sound like the latter: not “that won’t work,” but “here is what it would take to get as close as possible to what you have fallen for, and here is exactly where physics or planning draws the line.” And, dare I suggest, it’s helpful if this comes from a designer who is viewed as open-minded about AI, and not someone perceived to be an AI-sceptic.

At the critique.

This is the AI-as-amateur-friend problem, and in one sense it’s old as the hills. Clients have always had a well-meaning friend, relative or neighbour offering an opinion on the scheme. What is different is the register in which the opinion now arrives. The relative’s advice came pre-qualified by the listener: everyone knows to take Auntie Jean’s view on cabinetry with a pinch of salt. AI’s critique of your scheme carries no such qualifying tone, because the client never told it what the brief was, what the constraints were, or why the third option was rejected at concept stage. It is offering a confident opinion on a problem it was never given. The client, naturally, struggles to discount it the way they would discount Auntie Jean, because it does not sound like an amateur. It sounds like research.

At the contract.

The same mirage now reaches the paperwork. A client running your proposal or your terms through an AI tool will receive commentary that is fluent on contract law in the abstract and blind to the commercial context that makes a particular clause reasonable in your sector, at your fee level, for this kind of project. A little knowledge has always been a dangerous thing; what is new is how authoritative that little knowledge now sounds, and how quickly it can be generated.

Partway through the project, unexpectedly.

Perhaps the most consequential or devastating version of the mirage is the client who takes early-stage output, concludes they can “take it from here,” and uses AI to finish what you started. This really hurts when it constitutes IP theft: taking your creative input at a stage before design fees have caught up. This is the one example here of cynical (or worse) behaviour – and the type of challenge that kitchen designers will well recognise of old. Here the gap is not in the client’s understanding of design language, but in their understanding of everything you were holding that never made it onto the page: the trade relationships, the sequencing logic, the problems you could see coming that have not yet announced themselves. AI cannot see what was never written down, and the client, reasonably enough, does not know that it cannot. It’s time to start building in fee and contractual protection from this threat.

Why this matters for how we talk about it.

It would be easy to respond to all of this with irritation, and irritation is the wrong response, for two reasons. First, none of the above (well, perhaps with the exception of that final example) is the client’s fault in any meaningful sense; they are doing what any of us would do with a tool this fluent and this available.

Second, and more importantly for how we are perceived, a designer who responds defensively to an AI-mediated challenge confirms the exact suspicion that is the source of the challenge: that professional authority in this industry sometimes rests on the client simply not knowing enough to ask the awkward question. The better response is not defensiveness. It is the opposite: a calm, specific demonstration that you know something the AI does not, because you were in the room, you read the brief, you have seen this junction detail fail before. That is not a defence of your authority. It is a description of what your authority is actually made of.

And before we close the circle entirely around our clients, it is worth turning the mirage on ourselves. Every one of us, faced with our own accountant, solicitor, or surgeon, has likely felt the same small flicker of borrowed authority: a confident paragraph from an AI, primed and ready in our pocket, tempting us to believe we now know better than the person who has spent a career learning what we have spent an evening reading about. We're not exempt from the mirage, and hopefully seeing it turned on us, and holding space for other professionals in the same pinch, might go some way to reducing the impact over time. The honest discipline is not only to extend our clients more patience than irritation, but also to notice the same tug in ourselves, and to grant our own experts the same situational trust we ask our clients to grant us.

I'll leave suggestions for the practical handling of each of these scenarios for the newsletter. For now, the observation stands on its own: AI is not undermining the client relationship through any failure of the technology. It is undermining it by being exactly as good as it appears to be, and not one degree more situationally aware than that.

Sign up to receive the Hothouse newsletter here.