I want to share a story about how a computer architect, with the help of AI, started by solving a small problem of his own and gradually built a tool that more and more people wanted to use.

This week, after learning about the tool, a team asked me a very direct question:

“Why should we use your tool?”

I paused.

The question didn’t offend me. I’d simply never thought about it that way. I hadn’t set out to promote the tool, much less persuade anyone to use it. It started as a personal project to solve something that was frustrating me.

Now, people from different teams are choosing to use it. Some have even started contributing new features.

That has made me rethink how we create products in the age of AI.

It Started with a Painful Document

I’m a computer architect. But the first time I really felt AI’s potential to change how I work had nothing to do with hardware design.

I was reviewing a technical document with some very complicated tables. They were long, wide, and painful to read.

The usual process involved repeatedly moving between tools and document formats. Whenever the content changed, we might have to regenerate a file, open it, find the right place, and start reviewing again.

Oh god, my eyes hurt.

The most frustrating part was that none of this created much value, yet it took up a lot of our time. Experienced colleagues around me had been struggling with the same problem too.

I wondered: why couldn’t we put this content directly in a browser? Let the page adapt to the screen, make it easy to navigate and drag things around, and support interactions that fit how we actually work?

There was one problem: I had no web development experience.

At the University of Michigan, I’d once enrolled in a web development course, EECS 485. I dropped it after two weeks.

But by early 2026, AI coding agents had reached an interesting point. I decided to give it a try.

Over two or three days, while keeping up with my regular work, I used AI to implement and revise my ideas. I spent over a hundred dollars on tokens.

And a usable website came into existence.

It still had plenty of rough edges. But what stayed with me was a realization:

I didn’t have to master an unfamiliar technology before I could try to solve a problem.

Previously, I’d used the skills I already had to decide what I could and couldn’t build. AI made me start asking the question the other way around: if I set my technical background aside for a moment, what would be the best way to solve this problem?

That shift in thinking mattered far more to me than the website itself.

From a Personal Project to a Shared Asset

At first, I only expected a few colleagues nearby to use this little tool.

Later, we ran into an information management problem during our everyday collaboration. We eventually resolved it, but getting there was a hassle.

I thought: since we already had a better reading interface, why couldn’t we improve the collaboration experience too?

So I started adding features.

As we found more ways to use it, the tool kept evolving: from basic content display to collaboration, configuration, reliability, and eventually some AI-native features.

Many of these weren’t part of an original plan. The needs emerged through everyday use, and I addressed them a little at a time.

About six months later, the little website had become part of some colleagues’ daily work. More people from other teams began reaching out, hoping to use it to solve similar problems of their own.

But the moment the project really felt different was when I first saw other colleagues contributing new features.

I realized it was growing beyond something I’d built for myself.

Other people were willing to invest their own time in making it better. Some features came from use cases I’d never even considered.

A personal project was gradually becoming a shared asset.

That’s why the question this week—“Why should we use your tool?"—caught me a little off guard.

My answer was simple:

I don’t know whether you should use it.

I created the tool because I had a problem and wanted to solve it well. If you have a similar problem and find it helpful, you’re very welcome to use it, or even help improve it. I’m happy to support you.

But if it doesn’t solve your problem, there’s no reason to change how you work just to adopt a new tool.

I don’t believe in shoving a product down people’s throats.

I’d rather believe that a good tool earns its users by solving real problems.

The past six months have also taught me a few things about building products in the age of AI.

1. Start with a Problem, Not a Hammer

There’s a familiar saying:

When all you have is a hammer, everything looks like a nail.

I think it’s especially worth revisiting in the age of AI.

We each have a professional background, familiar tools, and methods we’re good at. When we face a problem, we naturally look for solutions within the abilities we already have.

I’m a computer architect, so I tend to think in terms of computer architecture. A software engineer might first think of writing a program. Someone skilled in Excel might reach for a spreadsheet.

These approaches are perfectly reasonable. The trouble is that the tools we already have can limit how we understand the problem itself.

AI is changing that.

What if we no longer have just a hammer, but a Swiss Army knife?

AI gives each of us access to capabilities we didn’t have before. We can explore unfamiliar fields, try different approaches, and build solutions we wouldn’t previously have considered.

With those options available, why keep looking for more nails to hit with the same old hammer?

I think one of the most valuable abilities in the age of AI is the willingness to reconsider your toolbox.

Being good at a technology doesn’t mean you should use it for every problem. Understand the problem first, then choose the tools.

That makes finding a worthwhile problem even more important.

AI can help us realize ideas that used to be difficult to implement. It doesn’t automatically tell us which problems deserve our attention.

Ideally, start with a problem you’ve experienced yourself. If you haven’t, you should at least have enough firsthand observation to understand what people are struggling with and why existing approaches aren’t serving them well.

AI expands what we can build. But we still need to figure out what is worth building.

2. Eat Your Own Dog Food

This is a principle I’ve tried hard to follow in this project and in other personal projects.

If I’m one of the tool’s intended users, I try to put it into real work from day one.

Even when the first version is imperfect.

Increasingly, I find that only by making a tool part of my daily work do I discover problems I never would have anticipated during design.

Some features look beautiful but feel awkward to use. Some designs are impressive in a demo but add friction in practice.

Meanwhile, small features I didn’t expect much from sometimes turn out to be surprisingly valuable.

This goes beyond the website. In other projects, I’ve often found that the way I initially imagined using a tool differed from the way it ultimately proved useful.

Sometimes, you have to build the tool and actually use it before you can begin to understand what it should become.

AI makes that process easier by shortening the distance between noticing a problem and trying a new solution.

Of course, my own experience can’t replace feedback from other users. But if a product’s creator isn’t willing to use it, it’s worth reconsidering what problem it actually solves.

3. Rethink the Workflow, Not Just Optimize It

This is the most important lesson I’ve taken from the whole experience.

Much of the discussion around AI still treats it as a way to optimize an existing workflow.

A process used to take ten steps; now AI automates three of them. Something used to take two hours; now it takes twenty minutes.

That’s useful, but I think it captures only a small part of AI’s potential.

AI should not merely optimize an existing workflow. It should make us question whether that workflow should exist in the first place.

Consider where my project began.

I just wanted to make technical documents easier to read. Following the conventional approach, I might have reduced the number of file conversions or written a script to automate more of the process.

But as I used AI to build a different kind of tool, I gradually realized that I wanted to change both how the documents were displayed and the entire workflow surrounding them.

If we designed that workflow from scratch today, knowing what AI can do, would we choose the same tools, the same processes, and the same division of work?

Probably not.

That’s why I don’t think AI-native should simply mean adding a few AI features to existing software.

To me, being AI-native means being willing to rethink how a problem should be solved from the beginning.

That doesn’t mean every old workflow needs to be discarded. Some have been refined over many years and still work very well.

But we shouldn’t assume a process remains the most sensible choice simply because it has existed for a long time.

Especially when AI has changed many of the assumptions behind it.

4. Iterate with AI. Constantly.

The last lesson comes from a product development experience that feels distinctive to this moment in AI.

In traditional software development, a feature might be out of reach because of technical limitations or because it costs too much to build. We usually accept the constraint and look for another approach.

Now, I think many of those judgments should come with an expiration date.

Early in this project, I tried to build some relatively complex interactions. The implementations AI produced had quite a few problems. After spending a lot of time debugging, I decided to set them aside.

A few months later, I tried the same features again.

To my surprise, they came together smoothly this time.

I hadn’t suddenly acquired a great deal of web development knowledge, and the requirements hadn’t fundamentally changed.

What had changed was AI’s capability.

That made me realize that something we couldn’t do a few months ago may be possible today.

More significantly, AI is reducing the cost of redesigning and refactoring existing systems as well as the cost of building new features.

That lets us revisit past design decisions more often.

We can try again with features we abandoned because they were too difficult to implement. We can reassess compromises we kept because changing them was too expensive.

Of course, this doesn’t mean constantly tearing systems apart just to use the latest AI technology. Stability, maintainability, and real users’ needs still matter.

But I’ve started to believe:

In the age of AI, “done” is becoming a temporary state.

A product’s current form reflects our imagination and needs, along with the technical capabilities and implementation costs at the time.

As those constraints change, we should be willing to reconsider what a better solution might look like.

Looking Back

What makes me happiest about this project is how it changed the way I look at problems.

In the past, faced with a frustrating but familiar workflow, I might have accepted it. The knowledge and time required to fix it often exceeded what I was willing to invest.

Now, I’m increasingly inclined to ask:

Why does it have to be this way?

With AI, the skills we already have no longer need to define which problems we can attempt.

And the fact that a way of working was once effective doesn’t mean it should last forever.

For me, perhaps AI’s greatest value is that it has prompted me to reconsider what work is worth doing and how it ought to be done.

The real opportunity of AI is not just to do things faster, but to rethink how things should be done.