A Block product designer spent 2 months vibe coding a dog ID app. Her top tip: Sometimes AI needs to be 'babied.'

Cynthia Chen vibe coded an app in two months with no formal training. She learned about how to better prompt AI — and that jobs like hers are safe.

• What started as a fun idea to collect dog photos turned into a full-fledged iOS app.
• A product designer with no formal engineering training built it from scratch through vibe-coding.
• Cynthia Chen shares with BI her process and takeaways from building Dog-e-dex.

When Block product designer Cynthia Chen first dreamed up the idea of an app that could catalogue dogs spotted in the wild, she shelved it.

"The hurdle was so big, so I never did anything," she said.

For five years, the dream sat untouched. Today, it's live on the App Store.

Dog-e-dex — a playful name inspired by Pokédex (not DOGE) — lets users snap a picture of a dog, identify the breed using image recognition, and add it to a personal collection. Chen has two cats and hopes to own a dog someday.

She built the app in about two months in her spare time with no formal engineering training — just pure vibe-coding.

"I have a lot of experience building products with other functions, like with engineers," she said. "But this is my first time actually building something myself."

Vibe coding, a term coined in February by OpenAI cofounder Andrej Karpathy, refers to the process of feeding prompts to AI to write code. As Karpathy puts it, developers can "fully give in to the vibes" and "forget the code even exists."

From dogs to code

Chen started with the app design in October and looked for a developer on Fiverr — an online marketplace for freelancers — but the cost was "intimidating." Then she realized she could use AI.

She turned to platforms like Replit, ChatGPT, and Cursor. It wasn't until she discovered Anthropic's Claude in January that things started to click.

She manually copied the code from Claude into Xcode — a tool for building apps on Apple devices — even when she didn't fully understand how it worked. "It was like magic."

"Every time I pressed the preview button, it was an exciting little gift opening," she added.

Building Dog-e-dex wasn't smooth sailing.

Her initial prompts correctly outlined the core functionality, but the app was "generating random dog breeds."

With a lot of trial and error, she figured out how to integrate an image recognition tool to detect actual breeds.

Along the way, she discovered that she didn't know what a backend was — the data and infrastructure that make an application work.

"Engineers would be like, 'What's your data schema? Where are your images stored?' And I was like, 'I have no idea, it just works,'" she said.

Chen said when she was stuck, she leaned on her engineering friends. But she credits much of the work to Claude.

If there's one thing she'd do differently, she would have started with a thorough feature list. Without one, she had to redo a lot of the foundational setup along the way, she said.

Prompting AI is like 'gentle parenting'

Chen has advice for people who want to vibe code: Treat prompting AI like "gentle parenting."

"You have to be very intentional, very specific, and I think you have to be very nice," she said.

One word matters: Changing a prompt from "Can you please build a navigation tab bar" to "Please build a custom navigation tab bar" can lead to very different results. The first generates a standard, built-in tab bar like the ones in iOS apps. The second creates a version with original style and behavior.

Sometimes, AI needs to be "babied," she added. When Claude got stuck, Chen would break down instructions step-by-step until it understood.

Design, she said, was "really hard to prompt," which gave her confidence that design jobs like hers are safe from AI disruption.

AI can't replace creativity — that's 'uniquely human'

Chen said the creativity needed to solve problems with AI feels "uniquely human."

While vibe-coding another app, she ran into a hiccup: The AI struggled to render a drawing at a small scale. She then exported the drawing as an image then resized it into a thumbnail.

"The AI didn't think of that, I thought of that," she said.

And while AI is trying to mimic personality — with varying results — it's humans who still think big picture.

"A computer can't just be random and funny, or think of a dog app," she said.

Chen thinks that what's holding people back isn't technical know-how.

"The limit is our imagination or our own sense of curiosity, our own sense of agency. It's no longer technical limitations," she said.

"Just try things. Don't worry about if you'll fail or not," she added.

Dog-e-dex is still in its early days, with about 75 downloads from the App Store.

Chen said she plans to begin marketing it soon. In the meantime, she finds joy in seeing users share the dogs they've collected and hearing how the app has added a bit of fun to their walks.

One user messaged her to say they collected their late dog in the app as a way to remember their furry friend.

"That was really touching," she said. "When it impacts someone's day and makes it better or fun or more delightful in any sort of capacity, that I think is the most heartwarming."