It's kind of a rite of passage for a new designer to get their first client and end up in "revision hell." Maybe the designer just jumped in and made something they thought was cool, but it doesn't connect to the client's vision, or perhaps they gave the client way too many options, so the client wants all of it, but all together. There's also that chance the client hates everything and wants to push every little pixel around until they come to an arbitrary point of satisfaction.
Experienced designers and agencies avoid this as much as possible, however, by going through a process of "discovery." This is where they do research, ask questions, put together workshops and / or make presentations on concepts. While the clients may find this long, it's actually really crucial because the designer needs this to do their best work and avoid endless revisions and frustration for both parties.
AI has started to do what seems like discovery by asking questions about requests for code or prompting clarification on complex questions. However, it still gets wrong what designers and agencies figured out a long, long time ago: never give a client a polished first draft.
The Illusion of Completeness
The advice to not give a solid initial piece may feel a little off-putting at first. Isn't it their job to deliver good work? Can't they just do what they're told and provide what the client wants as fast as possible? That's actually what AI does in contrast to what the best agencies and designers do, and I think the difference will help us understand what AI gets wrong, what it gets right, and what to be wary of when using AI. First, let's look at a proclamation in The Win Without Pitching Manifesto, by Blair Enns, to get a better look at what the problem is.
It is more likely that the client's perspective will be wrong, or at least incomplete, than it is that it will be whole and accurate. We know this. Doctors know the same of their patients. Lawyers and accountants know the same of their clients. The customer is not always right. More correctly, he usually has strong ideas and a strong sense that he is right, but is locked into a narrow view and weighed down by constraints that seem to him to be more immutable than they really are. When the client comes to us self-diagnosed, our mindset must be the same as the doctor hearing his patient tell him what type of surgery he wants performed before any discussion of symptoms or diagnoses. Our reaction must be, "You may be correct, but let's find out for sure."
When professionals pause, ask questions, and seek understanding we tend to get better results from our collaboration. The process should be one of transformation and refinement, which includes changing the minds of both parties involved as they work toward a solution. Designers have a particular trick that is necessary to slow down the frenzy and get to a better outcome: avoiding polish until the appropriate time. They do this because there's a bias that we have to get around as humans in which we assume something is complete if it looks complete.
This is the problem that AI is reintroducing—whether in a professional context or not. When you ask it to code something or to generate an image or write something, the AI does not produce a guided process with phases for check-ins and points of expert advice. Nor does AI challenge your own thinking like working through the problem with another person would. It skips straight to the end and pumps out a polished product. Sometimes agentic workflows can ask questions, but it's not just the act of asking questions that is important here. It's the process of working in collaboration with the expert that helps the client and the professional to come to a solid solution.
One of the advantages the outside expert brings is perspective. And one of the hallmarks of creativity is the ability to see problems differently, and thus find solutions others cannot see. To bring our perspective and problem-solving skills to bear we must be allowed time and freedom to diagnose the client's challenges in our own manner. Design is not the solution—it is the process.
The biggest traps we fall into when addressing complex projects or problems is the shortcut of assuming we've found the solution because the solution in front of us looks like it solves it. When AI provides you with exactly what you asked for, it looks complete, but it may not be. If you don't have expertise in the task you've asked the AI to perform, it's even more difficult to determine whether the product is complete, if it's safe, and if it's taking you in the right direction.
This is also why agencies in particular tend to have big presentation meetings. If you're rebranding or redesigning a website, the designers will often go through a thinking process that helps you understand why they made their decisions, why they pursued a certain direction, and how we came to the final result. (I did this on a small scale when I unveiled the CYBORG_ logo last year)
Even though these presentations are often more of a narrative explanation than a scientific journal of decisions made, it still helps introduce you to the solution and helps avoid the polarized reaction of love it or hate it. Especially for those who aren't in the field of expertise, the first time we see something new, we will have the most volatile reactions to it. The more you see something, the more it dulls the edge.
AI does not do this. It may flash its Chain of Thought notes as it reasons about your request, but it does not step you through each decision like an agency does. It rarely challenges your prompts. It summarizes its decisions and presents you with the polished product. That leaves everything to you to evaluate.
Shortcuts
I keep hearing warnings and advice about how we need to use AI but we also need to validate what the AI creates or does. I haven't heard much about how to actually do that, so perhaps some of the lessons designers have learned can help us here.
- Avoid going straight to the finished product. You may have to do the work of exploration on your own ahead of time and come to the AI to finish the task after you've made a decision and a detailed plan of action. This is very difficult. AI encourages us to shortcut to the end because it's so easy to just tell it to do something, and it always comes up with something that's close enough.
- Ask questions before, during, and after the production of a solution. Again, the friction to do this can discourage us from following through with this, because we are staring at a "finished" product. Even more difficult to ask questions and really investigate the decisions that went into the solution when you have little or no expertise to help you validate what you've been given.
Design is not the solution—it is the process.
You don't need the product at the end, you need the transformation of the process. Just because you're impressed that a few instructions to a computer produced something more than you expected, doesn't mean it's the right thing. You can only discover what's right during the process of working out what is right.
The real shortcut when using AI is to exercise restraint, which is an unfair burden placed upon users of an AI product that has been designed to be attractive for how it responds, reacts, and outputs what you've asked for. While we think we're getting huge value for very cheap labor, we're losing touch with our own skills, our own taste developed through expertise, and our own decision-making abilities.
Maybe the best shortcut is to avoid AI as much as possible when exploring something outside of your realm of knowledge and expertise, because you aren't going to be able to validate the product when it stands before you, shining with the illusion of completeness, polished with the hidden assumptions that you never articulated.