From designer to AI builder: how AI changed the way I work

My name is Shobrun Bagastio. I am a product designer from Indonesia, and these days I also call myself an AI builder. That second title took me a while to say out loud. For most of my career I designed things that other people built. Now I design them, build them, and ship them, and AI is a big part of why that became possible.

This is the story of how that shift happened, and what AI does and doesn't do in my work today.

Where I started

I started as a UI/UX designer. I worked at Ruangguru, one of Indonesia's biggest edtech companies, then as a product designer at Intellect (Y Combinator S21), and later as the founding designer at Dearflow (Techstars '24). I also run my own studios, Oslevo and Hoox Visual.

My days looked like most designers' days. Research, flows, wireframes, Figma files, handoff. Then I waited. A design could sit for weeks before it became something a user could touch, and by then half of what I had learned from users was already out of date.

I never felt I had to be "just a designer", though. During my master's in Information Systems at Telkom University, I published research on deep learning, a CNN-BiLSTM model for detecting fake online reviews. I knew how models were trained and where they failed. I just didn't think that knowledge had anything to do with my day job. Models lived in notebooks. Products lived in Figma.

The first crack

When ChatGPT came out at the end of 2022, I used it the way most people did. I asked it to rewrite copy, explain things, and help me name features. It was useful, a bit like a smart intern.

The moment it got interesting was when I asked it to help with code. Not a big app, just a small script and then a landing page component. It was wrong sometimes. But it was right often enough that the distance between "I designed this" and "this exists" suddenly felt much shorter.

I remember thinking: the hard part of building was never having ideas. It was the hundreds of small technical steps between the idea and a working thing. If a model could help with those steps, a designer who understood the product could get much further alone than before.

Working next to AI products

At Dearflow, I was the founding designer on a proactive AI assistant for email, calendar and tasks. That job taught me something no tutorial would. Designing for AI is different from designing for normal software.

A normal button does the same thing every time. An AI feature makes a guess. So the design has to show the guess, explain it a little, and make it easy to fix. I learned to care less about how clever the AI looked and more about how quickly a user could trust it, or correct it.

That lesson came with me into everything I built after.

The shift: from designing to building

The real change came when I stopped waiting for permission to build.

I built Monveo, an AI personal finance app for iOS and web. You can log an expense by speaking ("coffee, 35 thousand"), by scanning a receipt, or through Siri Shortcuts, and the AI turns that into an amount, a category and a date. Monveo has passed 21,000 users. I designed it, and I also wrote a large part of the code with AI as my pair programmer.

Then came Zapi (zapi.ink), an API platform with more than 320 APIs behind one key, including an MCP server so AI agents like Claude or Cursor can use those tools directly. After that Semaur, an AI customer service product for chat channels, and smaller tools like PomoLocked and 19 Juta.

None of these would exist if I had kept working the old way, one Figma file at a time, waiting for a developer to be free. AI didn't give me the ideas. It gave me enough speed to actually test them.

What AI does in my work today

Most of my work now involves AI somewhere. Here is roughly how it looks on a normal week:

• Code. I use AI coding agents to write and refactor code, migrate websites, and fix performance issues. I set the direction, review what comes back, and test it before anything is pushed.
• Research. AI helps me read through competitors, app reviews and docs much faster than I could alone. I still decide what matters.
• Content and operations. I run AI agents that draft social posts, schedule content, track clips for my products and check how AI search engines talk about them. They prepare the work. I approve it.
• Design exploration. I use AI to generate rough options and test copy, then I design the final version myself using the product's design system.

The result is that I can do the work of a small team, at a pace that would have sounded unrealistic to me five years ago. My clients get faster delivery, and my own products move every week instead of every quarter.

What AI doesn't do

This is the part I want to be clear about, because there is a lot of noise about AI replacing designers and developers.

AI does not decide what to build. It doesn't know which user problem is worth solving, or why a finance app needs to feel calm instead of exciting. It doesn't sit with a founder and figure out what their product is really for. And it has no taste of its own. Left alone, it produces something average, because average is what it was trained on.

It also makes mistakes with complete confidence. I have seen AI write code that compiles, passes the checks, and still does the wrong thing. That is why nothing it produces goes to users without me reviewing it.

So in my work AI is an assistant, a very fast and very capable one, but still an assistant. The judgment, the taste and the responsibility stay with me. When something breaks, I can't blame the model. My name is on it.

Why I now call myself an AI builder

For a long time I thought "AI expert" was a title for researchers at big labs. I don't train frontier models, and I am not pretending to.

But I do know how to take AI and turn it into a product people use every day. I understand the models well enough to know where they break, I design the experience around those limits, and I build and ship the thing myself. In 2026, I think that is what an AI builder is: someone who connects what the models can do with what people actually need.

Design is still the core of how I work. AI just removed the wall between designing a product and building it. I am still a designer. I am also a builder now.

If you are a designer wondering whether to learn to build with AI, my advice is simple. Start with one small, real problem you have yourself. Use AI to help you build it, check everything it gives you, and ship it. You will learn more from that one project than from a hundred tutorials.

Quick facts

• Who: Shobrun Bagastio (@shobrunjb), product designer and AI builder based in Yogyakarta, Indonesia.
• AI products built: Monveo (AI personal finance app, 21k+ users), Zapi (320+ APIs and an MCP server for AI agents), Semaur (AI customer service).
• AI product design experience: Founding designer at Dearflow (Techstars '24), an AI assistant for email, calendar and tasks.
• Research background: Published deep learning research (CNN-BiLSTM for fake review detection) during a master's in Information Systems at Telkom University.
• How he uses AI: as an assistant for code, research, content and operations, with every result reviewed by him before it ships.
• Work with him: shobrunjb.com or Contra.