I came to AI as a designer, and that is why my builds stick
Engineers can make AI work. Designers make it usable. The gap between those two is where most AI projects quietly die, somewhere between it technically functions and a real person would actually use this.
Plenty of AI projects work and still fail. The thing does what it was built to do, the demo goes fine, and then nobody uses it. A few weeks later it is quietly switched off.
If you have watched that happen, you already know the problem was never that it did not function. It functioned perfectly. It just was not usable.
I came to AI as a designer and developer, and I think that is why my builds tend to stick. The functioning was never the hard part.
Engineers make it work. Designers make it usable.
These are two different skills, and most AI gets built with only the first one.
Making it work is getting the model to produce the right output. Real, important, and only half the job. Making it usable is everything between that output and a busy person on a Tuesday who has thirty seconds and no patience for a clever tool that is hard to use.
The gap between those two is where most AI projects die. Somewhere between it technically functions and a real person would actually choose to use this.
Good AI feels like good design
The best AI I have built is the kind you barely notice. It does not announce itself. It does not make you learn its quirks. It quietly does the thing and gets out of your way.
That is just good design wearing a new coat. Good design is invisible. You notice bad design constantly and good design almost never, because good design simply works and lets you get on.
AI is no different. If your people can feel the AI, if they have to think about how to handle it, it is getting in the way. The aim is for it to disappear into the work.
Designing AI well is mostly about reducing friction the user never sees. You give it a tight system prompt so it stays in role and does not need babysitting. You ground it in the right data through retrieval, often called retrieval augmented generation or RAG, so a person never has to feed it context by hand. You add a validation layer so wrong or off brand output is caught before anyone reads it, which means the person trusts what they get rather than checking every line. You decide the one moment a human is asked to step in, and you make that moment obvious and quick. Each of those is an engineering decision, but the reason for each one is a design decision: protect the user's attention and earn their trust. Good AI is good design, and the work is making all of that effort vanish from view.
Why the designer's eye makes builds stick
A build sticks when using it is easier than not using it. That is the whole test. If the AI saves real effort and asks for almost none in return, people keep it. If it saves effort but demands fiddling and second guessing, they drift back to the old way.
That calculation is a design problem, not an engineering one. It is about attention, trust and habit. It is about how the thing feels to use at the moment someone is tired and in a hurry.
I work alongside these tools every day, and the builds that last are not the cleverest. They are the ones that respected the person on the other end.
The part most projects miss
So if you have an AI tool that works on paper and nobody touches, the missing piece is probably not more engineering. It is design. It is the unglamorous work of making a capable thing genuinely pleasant to rely on.
Make it usable and it sticks. Leave it merely functional and it joins the quiet pile of clever tools that nobody opens any more.
If you have AI that works and nobody uses, the missing piece is usually design, not engineering. Making capable tools genuinely usable is the part we care about most.
Book a quick chat →Related: I design systems, not prompts.
Common questions
Why do AI tools get built and then never used?
Usually because they were made to function but not to be usable. The model produces the right output, but using the tool asks too much of a busy person, so they drift back to the old way. That gap between technically works and a real person would choose this is where most AI projects quietly die.
What does it mean for AI to feel like good design?
It means the AI is almost invisible. It does not make you learn its quirks or babysit it. It quietly does the thing and gets out of your way, the same way good design simply works and lets you get on without noticing it.
Is making AI usable an engineering or a design problem?
Both, but the reasons are design ones. Tight instructions, grounding in the right data, validation and a clear human checkpoint are engineering decisions, and each exists to protect the user's attention and earn their trust. The aim is to make all that effort disappear from view.
Why do design led AI builds last longer?
Because a build sticks when using it is easier than not using it. That is a question of attention, trust and habit, not cleverness. Builds that respect the person on the other end get kept. Builds that demand fiddling get abandoned.