Around half of the design briefs reaching D2M today are what our Head of Design, Hamit, calls “AI reasoned” documents: briefs where the client shares a rough idea and lets AI fill in all the gaps. They look complete and highly technical, but they often skip the most important questions, such as who the product is for and how it will be used. In this article, Hamit shares real examples from our desks, explains what these briefs cost both founders and agencies, and sets out how to use AI properly without letting it do the thinking for you.
Last week, we received a comprehensive design brief along with a design enquiry. A complete document of more than ten pages, packed with in-depth technical detail about the proposed product idea, right down to the battery requirements. After spending an hour reading through and assessing the details, we came to the end, where a big disclaimer asked us to ignore everything listed in the document, as it was completely AI generated and only there as a “suggestion”. The ultimate frustration, after an hour of taking notes and thinking it through.
I do have to say that this was not a one-off enquiry. It has become a frequent kind of application landing on our desks recently, and I’ve decided to name them “AI reasoned” documents.
Let me explain what I mean.
From napkin drawings to 40+ pages
Just like any design consultancy, D2M accepts enquiries through a form so we can assess them efficiently. After the form, an NDA is put in place and we start talking with our prospective clients. During these early conversations, we usually receive the depths of the idea we are about to pursue, in all sorts of mediums.
Until a year ago, this could be anything from a sketch on a napkin (yes, it happened) to a proper design brief with all the questions addressed. Honestly, we appreciate all of these approaches. Yes, things were a bit tough to work through at first, and we had to ask for a lot of clarification, which resulted in more meetings... but the thinking behind them belonged entirely to the person who sent them.
Today, things look very different. Around 60 to 70% of the briefs we receive contain AI generated imagery, and around half of all the documents we receive are written as what I call AI reasoned documents.
To be fair, AI has improved the overall quality of what we receive, and we are really pleased with that. But it also brought its own complexity.
The AI reasoned documents
An AI reasoned document is born like this. A client has a general idea. Rather than thinking it through fully, they lean on an AI agent and expect it to fill the gaps. Technical gaps, definitions, specifications, all of it. And if you let a capable AI agent run with it, it will happily fill in every detail as if the product is ready to go into manufacture.
The result arrives in 20, 30, sometimes 40 pages. Charts, performance planning, component lists, manufacturing level specifications. It looks complete. It reads complete. But the simple foundational design decisions were never taken in the first place, which makes much of the document obsolete before we even begin.
Locked and open at the same time
Back to last week’s brief.
It specified a certain type of product at a nearly technical level. Which light, which lens, which speaker, what type of battery, how it should be stored and sourced, what colour the product should be, what material it should be made of. All defined, one after another, as if it were going to the factory tomorrow.
What we could not find anywhere in those ten pages was the user. Yes... The document lacked any detail on who this product is for. How will people interact with it? Where will it be used and stored? What about the interaction, the engagement? These key questions set the parameters of the design. They define the size, material, structure, durability, weight and colours of the proposed product. The user was kept as broad as possible, which stopped us from imagining any scenario, yet the part thicknesses and materials were already defined!
In short, the big existential questions that are a MUST for any value proposition weren’t answered at all, yet the specs were defined down to the last screw size. It was as if the user details and all the necessary questions had been answered somewhere and were simply hidden in the document, because it spoke as though everything was already defined, from colour to every other key detail of the product. There was zero room left for design to explore, and zero thought process visible about the people who would actually use the product.
While we were battling with hundreds of thoughts, along came the disclaimer. Everything is AI generated. These are idea suggestions only. None of the measurements are realistic, none of the technical details are realistic, the battery and electronics are not finalised, and the designer should change all of it.
So, all borders are open now... except they are not.
The client has already put something out there. There is an expectation, a definition, a thought process (even if it isn’t entirely theirs), and as an agency we respect that. You cannot simply unread ten pages. The design borders are locked and open at the same time, and that is exactly where the gridlock begins.
Imagery carries the same problem. AI will generate whatever image you ask for, but it doesn’t do a reality check. It doesn’t know whether a product can be manufactured. Yet people see a convincing image and assume it can be built and sold. (We are already seeing plenty of AI generated products being sold online that simply don’t exist the way they are shown.)
I would also like to share a few additional examples of requests we have received in the past couple of months, and I can assure you these were real, down to the exact details.
An AI generated image with a few dimensions, and a request to reverse engineer it, render it in perspective and develop the product. A product that was never engineered in the first place.
And my favourite, if I can call it that... a business asking us for prototyping and product development on a design they had already spent £10,000 patenting. The design was entirely AI generated. The AI agent had told them the idea was unique and should be patented, and even crafted the patent information for them to file. Which they did. In our very first call, we found a couple of existing patents covering similar ground, and the product would not work the way they envisioned. Ten thousand pounds spent on a patent that is unlikely to hold up, for a product that doesn’t work.
What happens on our side
When an enquiry comes in that we believe we can genuinely help with, we reach out and ask for more details once the NDA is signed. This is usually the point where the AI reasoned documents arrive.
What follows is a quick internal session. Can we develop this? Can we provide our services efficiently? Does it suit our skills and abilities? Can we really contribute to this project? We don’t want to waste anyone’s time, and that is one of our company characteristics.
We don’t automate this step and we don’t run it through an AI agent. Every project is evaluated directly by a person, because we want to understand what quality and capability we can bring to make the project commercially viable and create a positive outcome. That assessment alone takes a couple of hours.
Then comes the quotation. D2M quotes stage by stage. We plan out each stage and price it realistically rather than summing up one broad number, because every project has a budget that matters and we like to stick to it properly. That takes time and consideration too.
With an AI reasoned brief, we sit back and say, wait a minute. What kind of design service are they actually looking for? Is it manufacturing? Which part of this heavily defined design do they want explored or revised? Which details are fixed and which are just suggestions?
When we get caught up in this back and forth, on/off type of questioning, and some details start to challenge the rest of the project or the very foundation of the design scope, we have to pause and get back to the client with further questions.
Of course, we can’t say, “Hey, is there a possibility that you used AI to answer these details?” Instead, we ask how much of the detail shared with us is locked in and vital to them. If we find a contradiction, a material that isn’t feasible to manufacture or a battery that won’t be enough for what the product needs to do, we point it out and ask how they would like to move forward.
This usually turns into another call. Then often another one. Overall it extends the process by around 50% before we have even written a proposal.
And the most frustrating moment? When we challenge one of these points and the client steps back and says, “Oh, we didn’t think about that. It was just done with the prompt.”
That single sentence undermines the whole foundation of the brief, immediately.
Scope is everything for us. As we always say, design is infinite. You can keep designing a product back to back forever and never finish. But everyone has a budget and everyone wants a certain function out of a project. So the features, the user, the who, what, where and when all need to be defined. When those foundations are missing, we have to quote on the safer side to compensate for the time lost and the clarifications we know will come further down the line. It is never a nice situation, for us or for the client.
The risk on the client’s side
Just like manufacturing quotations, where three factories can come back with completely different prices, three agencies can do the same. There is no standard rate here, as there is when you see a doctor or a lawyer. Product development is expensive, and the price depends on each agency’s capability, hourly rates and understanding of the problem.
Now add an AI reasoned document into the mix.
A meticulous agency will read it carefully, ask a lot of questions and quote higher to stay on the safe side. Another agency might quote much lower without reading it properly, missing or ignoring the scope just to win the work. And since the client never fully reasoned the project themselves, they don’t really know what is inside it either. An agency can bend those details to its own leverage, and the client ends up with a completely different product from the one in their mind.
In the worst case, a different and poor outcome. In the best case, a much more expensive quote.
It is also worth knowing that agencies come in all sizes and approaches. Some love to deliver concepts but don’t want to get into the hard work of prototyping and manufacturing specifications. With an AI reasoned brief, these agencies will often quote low, deliver the concept, and leave the project once things get serious, usually right when you need a Mark 1 prototype to prove the design actually works. Not everything is crystal clear on a website, so it pays to ask.
For us at D2M, this simply isn’t an option. We cannot release a concept that isn’t feasible to manufacture and commercially viable. If something can be made but we believe there is no market for it, we flag it to the client straight away. So we have to fact check every detail, turn every stone, understand the client and the direction of the project, lock it down, and only then move into development.
So, how should we use AI?
Let me be clear. I’m not against AI. We are living in the age of artificial intelligence and using it is becoming unavoidable. Personally, I think it is an important instrument to fact check your thoughts, ideas and thinking, to research and audit, and to learn more about the market you want to develop your product for.
But there is one rule I would ask every founder to follow.
Never let AI reason your product.
Do your own thinking first. Before opening any AI tool, answer these yourself. Who is this product for? What is it made for? What features will it have, and will those features be valuable to the segment you are developing for? Is it feasible? Can it survive in the market against what is already out there? Where will it be used, and how will it be maintained?
These answers show your thinking and your thought process. Once you have them, even as a rough template, use AI to run market research, do patent research, fact check your idea and find alternative options. Ask it for multiple options, then validate them and build your own synthesis.
Keep in mind that if you simply ask an AI agent, “Is there a market for this product?”, it is designed to please you. It will most likely say yes and politely appeal to what you want to hear... unless you specifically ask it to challenge, contradict or critique your idea.
And it needs proper data in the first place. Without the foundations laid out, there is simply not enough for it to work with. Good thinking in, useful detail out.
Interestingly, this is the same thing I used to stress to my design students. Let them run the research, generate the insights and define the problem themselves, form the scope, and only then start ideation. Not the other way round. Founders deserve the same discipline, AI or no AI.
To wrap up
AI has made the briefs we receive look better than ever. That part is real, and I’m glad of it. But a brief that looks complete is not the same as a project that has been thought through.
If you are about to send your idea to a design agency, take some time with the who, what, where and how first. It will save you time, money and quite possibly ten thousand pounds.
And it will help us help you, much faster.
Best,
Hamit
