Useful before impressive: Product design in an AI world
When I think about a product I want to build, I try to start with the moment someone actually needs it.
A parent is trying to understand what a school message requires. A student wants to share something they have been working on. A teacher needs to complete a task before the next class begins. None of them arrived hoping to admire the number of features I managed to include.
They have something to do. My job is to make that possible without adding unnecessary work.
That is what I mean by utility-first design. It is an order of priorities, not a particular visual style. Before I decide how a product should present itself, I want to understand what it should make easier and for whom.
Starting with the actual task
A feature can sound useful without fitting the situation in which someone would use it.
Consider a school update. From the product side, the task might look like sending a notification. From the parent's side, it could involve understanding the message, deciding whether action is required, finding the right form, and knowing whether the response reached the right person.
A successful notification is not necessarily a successful experience. The whole sequence matters.
This is where I want to spend more of my attention. Can the person tell what is happening? Is the next step clear? Can they correct a mistake? When they finish, do they know they are finished?
These questions seem ordinary, which is part of their appeal. They bring a product back to the circumstances it is supposed to serve. They also make it harder to confuse a long list of capabilities with a well-considered experience.
I would rather make a necessary task reliably manageable than add a feature mainly because it looks good in a demonstration.
Accessibility is part of whether it works
Usefulness cannot be defined only around someone who uses a product exactly as I do.
The World Wide Web Consortium describes web accessibility in terms of people with disabilities being able to perceive, understand, navigate, interact with, and contribute to the web. That includes how a product is coded, not just how it looks. Keyboard operation and compatibility with assistive technologies are part of the work.1
This matters to my understanding of utility. If a student cannot operate a control, the control is not useful to that student. If someone cannot understand the instructions, a technically functioning form still has not done its job.
Minimalism alone does not solve this. Removing visible labels might make a screen look cleaner while making it harder to understand. A pale color palette can be visually restrained and still difficult to read. A single gesture can be elegant for one person and unavailable to another.
W3C also distinguishes accessibility from general usability. The two overlap, but a product that feels easy to use for some people can still exclude people with disabilities. Both need deliberate attention.2
For me, that means keeping the people involved in view throughout development. It also means treating accessibility checks and feedback from users as part of building, rather than a final statement to add to the website. I want the work to earn the claim.
What AI changes for me
AI makes this priority feel more important, not less.
Generative systems can help produce text, images, and code. That expands what can be drafted or prototyped, but the ability to produce an output does not settle whether it belongs in a product.3
My concern is that greater building capacity can turn into greater product clutter. If I can add another interaction, another dashboard, or another assistant, it becomes tempting to treat the addition itself as progress.
I would rather use the capacity to remove work from the person using the system.
Sometimes that could involve AI. It might help someone make sense of a long message or express a question in their own words. Sometimes it could mean a clearly labeled button and a dependable form. I do not think a person should have to compose a prompt to complete a task that the product could simply make understandable.
There are also moments when automation needs an obvious boundary. A draft should be recognizable as a draft. A consequential action should have an appropriate confirmation. A person should be able to reach a human when the automated route is not helping.
The design question is not how prominently intelligence can be displayed. It is where assistance earns its place and what happens when it is wrong.
Useful does not have to mean dull
I care about visual design. I like products that feel considered, distinctive, and enjoyable. Utility-first is not an argument for removing those qualities.
A clear visual hierarchy can help someone understand a page. Warmth can make an unfamiliar tool feel less intimidating. A well-chosen illustration or a small moment of personality can make an experience more pleasant. There is room for play, particularly in products involving students.
I simply want those choices to support the purpose of the product.
The same is true of marketing. People need a way to discover something, understand it, and decide whether to trust it. A useful product does not automatically find its audience. Communication matters, and so does distribution.
What I want to avoid is asking presentation to carry a promise the experience cannot keep. If someone understands the product better after reading the website but feels misled after using it, the gap deserves attention.
This is one place my decision framework becomes practical. The A1 choice might include excellent design and thoughtful marketing. It just cannot use them as substitutes for solving the problem.
Choosing what to measure
I also want the measures of success to follow the purpose.
For a tool designed to help someone complete a task, more time spent inside it may not be an improvement. Fewer steps might help, but not if the missing step protected someone from a costly mistake. More notifications might increase activity while making the underlying relationship harder to manage.
I would want to know whether people completed the task, understood what happened, and needed less help to do it again. I would also want to know who could not complete it and whether the apparent improvement hid extra work for somebody else.
These are the standards I am trying to bring to Education Angel Group and to my other projects. They do not answer every design question, and applying them consistently is work of its own.
But they give me a place to start. Before making something more impressive, I can ask whether it has become more useful to the person who came there for a reason.
Sources and notes
- World Wide Web Consortium, Web Accessibility Initiative. Introduction to Web Accessibility. ↩
- World Wide Web Consortium, Web Accessibility Initiative. Accessibility, Usability, and Inclusion. ↩
- NVIDIA. What Is Generative AI?. ↩