In the right hands

What to do when your client turns up with the design already done - by AI.

A garden designer told me about a consultation that changed shape before she had taken her coat off. The client met her at the door with a folder of renders and said, 'I don't need your design skills. I did the design with AI. I just need you to plant it.'

The design was striking. It was also unworkable. Plants that would never survive that soil, a layout with no sense of how anything drains or grows, species that will swamp the borders inside three years. The client could not see any of that, and there was no reason they should - it looked finished, and it had cost them an evening rather than a fee.

She is not the only one having this conversation. A recruiter I know watched a client switch to an AI screening tool that quietly rejects the best candidates it sees - nobody notices, because nobody meets the people it filtered out. A business we spoke to built its own reporting solution and cancelled the specialist who used to do it; the reports look credible and are wrong in places that matter. And there is a blunter version arriving now: 'we're doing it all ourselves - we don't need you any more', from clients who have built their own inventory control or reporting software without a supplier at all.

If some version of this has happened to you this year, you are not imagining the pattern, and it is worth saying plainly why it is happening now. The latest generation of AI models - Claude 4.6 and its peers - crossed a line in the last few months: a motivated person with no training can now produce something that looks finished in an afternoon. Not something that is finished. Something that looks it. That distinction is the subject of this blog.

I run a software business, which means my trade met this moment about six months before yours did. Prospects began arriving with AI-built prototypes and asking us to 'just make it production-ready'. I remember the chill. So what follows is not commentary on a trend; it is a report from the far side of a moment most professions are having right now - what we got wrong, what we learned, and the model that works.

Your customer is half right

Start with the uncomfortable part: the client in each of those stories has a point.

Their costs are rising on every line and professional fees are one of the few they feel they can control. The tools they are using are astonishing - options in seconds, tireless iteration, visualisations that once cost thousands. If you have not used them yourself, you are underestimating them; the power is real and it deserves to be celebrated rather than resented.

And here is the part that matters for what follows: at proposal stage, your customer cannot tell the difference between finished-looking and finished. The gloss is identical. Knowing the difference is not their job - it has always been yours. That is precisely why they used to hire you, and it is why the way you respond to this moment matters more than the moment itself.

So this is not a story about foolish clients. They are behaving rationally in a squeeze, with real power in their hands for the first time. The question is what happens next.

When they find out

What the gloss hides tends to surface in a predictable order.

The brief was wrong. AI answers the question exactly as asked, and asking the right question is the hard part. The garden client asked for a Mediterranean scheme; the site is heavy clay in a frost pocket. A professional interrogates the brief before answering it - it is half the value, and the half nobody sees on a render.

The generalist gap. Today's AI is trained on the whole internet, which makes it the most capable generalist ever built - and a generalist is what it remains until someone teaches it a particular world. It has read everything in general and knows nothing in particular about this soil, this data estate, this labour market, this planning authority. The missing ingredient is not intelligence; it is context, and context is what a professional has spent years acquiring.

Day one is not year two. The self-built reporting tool works at launch and drifts quietly out of truth as the data changes underneath it. The screening tool runs beautifully while rejecting the person who would have transformed the business. The AI-designed garden looks wonderful in July and is a bog by February. Convincing wrongness is the dangerous failure mode - obvious rubbish gets caught at the door.

The 2am question. When it fails - and complex things fail - who takes the call? Who is insured, accountable, obliged to put it right? An AI subscription carries none of those duties. A professional carries all of them, priced in.

Software has already run this experiment at scale. This year we have seen a wave of AI-built prototypes that could not become products - unmaintainable, insecure, unfixable by the people who generated them - and the repair usually costs more than building it properly would have. The lesson was not 'avoid AI'. We use it every day. The lesson was that generated and engineered are different things, and the difference is a person who knows what right looks like.

The losing response

A design studio I heard about was shown a client's AI-generated scheme and told them, accurately, that it was poor. They were right, and they lost the job - the client simply found someone less insulting.

Refusal, bans and mockery all fail the same way, because they misread what is standing in front of you. The client who arrives with an AI design is a qualified, motivated lead. They have shown you exactly what they want, they have invested their own time in wanting it, and they are ready to act. Laughing at their homework converts none of that.

There is also a competitive clock running. The professional next door who adopts these tools will match your quality and beat you on speed and price. Your customers will not choose between you and AI; they will choose between you-without-AI and someone-like-you-with-it.

And the client who left entirely - the 'we don't need you any more' one? A fair share of them will be back when the year-two costs arrive, quietly, with a system nobody can maintain. Whether they come back to you depends entirely on how you behaved on the way out. Burn no bridges; the tide comes back in.

The model that works: 1 + 1 + 1 = 4

The best project outcomes I have seen this year had three parties in the room, each doing what they are best at.

The client brings goals, taste, budget truth - and, increasingly, preferences they have already explored with AI. Welcome that folder of renders. It is the best-briefed a client has ever been, and it tells you exactly what they are hoping for.

The professional brings judgement, context and accountability - the interrogated brief, the years of knowing what this soil or this dataset or this market actually does, and a name on the work when it matters. Crucially, the professional also leads the AI.

The AI brings speed, options and tirelessness - in the professional's hands, working from the professional's context: their portfolio, their standards, their regulations, their site knowledge, their data. Taught that world, the great generalist becomes a specialist, and it is worth several times more than the internet-trained version the client used alone.

Replay the garden consultation in this model. The designer says: 'This is a lovely starting point - you have told me exactly what you are drawn to. Now let me show you what it becomes when we teach the same tools about your soil, your light and how this garden should look in five years, not just next month.' The client gets a better garden, faster and at lower cost than the old way. The designer keeps the relationship, the fee and the craft. The work is better than either could have produced alone. One plus one plus one makes four.

The unavoidable line underneath all of this: lead the AI in your practice, or your client will lead it for you.

We did it to ourselves first

This isn’t one of those cases where a consultant points out what you should be doing from a position of... well, being a consultant.

three rocks has been building software for 25 years, and ours was one of the first professions this happened to - code was commoditised ahead of garden design, building plans and candidate shortlisting. When clients started arriving with AI-built 'finished products', we made a decision that felt risky but turned out to be the right answer: we rebuilt our own practice around the tools instead of arguing with them.

We did this as a team, not as individuals experimenting quietly. A working group investigated which tools and techniques actually fit our workflow - not which ones demoed well. The hardest part was human, not technical: our developers were sceptical, and rightly so - they were being asked to trust this new technology to do work that they would be held accountable for. Remember the first time you saw a car park itself?

That scepticism proved to be our greatest asset. No one got too giddy making fun AI pictures or slick-looking product presentations that didn’t quite make sense under scrutiny.. We put the most hard-to-impress developers (there isn’t really another type) to work testing and adapting the tools to our specific standards.

The result is a growing set of skills, plugins and hooks that make handing work to AI safe and repeatable. Our AI tools aren’t just the out-of-the box AI solutions that know everything in the internet but not much about our business and goals - they arrive already knowing our standards, our patterns and our definition of ‘done’.

We deliver faster and better than we did two years ago, with the same people carrying the same accountability. Having made this change, we’re ready for – and excited about – the next generations of AI that are coming. We’re in that world now, there’s no going back.

AI tools did not do this for us. It was our software engineering experts’ discipline and focus in developing the tools to be right for our kind of work.

That experience - taking a professional practice and putting AI inside it, without losing the judgement that makes it a profession - is now something we build for others.

An invitation

If you are a professional watching this pattern arrive in your consultations, the answer is neither surrender nor refusal. It is leadership - of the tools, and of the three-way relationship between you, your clients and AI. If this essay has landed close to home, I would like to hear your version of the story: get in touch.