Exploring funded brick-and-mortar companies on Wefunder isn’t easy, so I asked Replit Agent to build a map and see how much time AI could save me versus doing it manually.
In about 3 hours and for roughly $22, it produced a fully searchable map of funded companies. Most of the cost came from the Agent inefficiently debugging an API integration and unnecessary reloads. If I had built it myself, it probably would’ve taken 4–5 hours to design, build, test, and deploy.
Check out the code or explore the map below:
Repo: 🔗 Wefunder Brick & Mortar App by kenakingkong on Replit
Demo: 🔗 wefunder-brick-and-mortar-map.replit.app
Keep reading for a full breakdown of what worked, what didn’t, and where AI saved time versus where it didn’t.
Background
Wefunder has thousands of funded companies, but the only way to browse them is by scrolling through media-heavy cards on the Explore page. (eg: wefunder.com/explore/brick_and_mortar).
There is a map feature, but it only shows companies that are currently raising. I needed a better way to explore funded brick-and-mortar companies.
Replit launched Replit Agent in 2024, an AI agent for automating software development and building web apps.
I’m a full-stack engineer with 5+ years of experience (makenakong.com/dev), so normally I’d just build this myself. But with all the hype around vibe-coding, this felt like a low-stakes project to test it out.
Building the app
Getting started
My first prompt was:
“Build me an interactive map of all funded brick-and-mortar companies on Wefunder.”
The Agent started looking for a public Wefunder API, which exists but is not publicly accessible. I should’ve told it upfront that I had a private API endpoint ready.
After a few minutes, I updated the plan with the correct endpoint and example JSON. Another minute later, I approved the revised plan and hit Build the entire app.
Eventually, the app loaded, but without any companies. And that’s when things got hands-on.
“Development”
Setup and Initial Structure
The agent installed the packages and generated core components, including Leaflet, an open-source JavaScript library for interactive maps. I’m unfamiliar with the package so it would’ve taken me a while to read the documentation and render it properly.
API Endpoint Issues
First, I tried using the Explore page endpoint, but hit a CORS wall.
Then I generated an external API key and tried using Wefunder’s private external API (the one KingsCrowd uses).
The JSON structure was totally different.
It wasn’t filterable by category.
This confused the AI and resulted in non–brick-and-mortar companies showing up.
After watching the agent repeatedly debug and regenerate with no progress, I finally parsed the endpoint explore page data myself and hardcoded it.
This would’ve gone much smoother if Wefunder had a clean, public API. And, this would’ve been faster if I had parsed the JSON response and debugged it myself as well.
Fine Tuning
Once the data displayed correctly, I fine-tuned the UI/UX through small prompts, like:
- Remove all icons and make the total raised and investors inline
- Add a “funded ” to the total raised and investors section
- Remove search by category functionality from the sidebar
- I asked it to fix the position of the search icon in the search box and it took 8 minutes and cost $1.38.
Each request took between 1–8 minutes and cost $0.30-$2.00.
Realistically, I could’ve done most of these edits manually faster and for free.
Hosting
The agent published and hosted the site directly. It failed once, then succeeded.
Other issues
- Replit Agent sometimes removed only parts of a feature instead of all of it, leading to compilation errors.
- It often reloaded the entire app after tiny changes, instead of doing partial reloads, burning extra credits.
The search functionality still needs work but I’m running low on credits.
Comparison
What I Liked
- It saved time on setup and boilerplate.
- The default UI was perfectly fine for a simple app.
What I Didn’t Like
- Watching it reload endlessly.
- Watching it chase the wrong debugging paths.
- Watching my credits evaporate when I could’ve fixed things faster myself.
- Not knowing what the code looked like at every moment.
Replit AI vs Me: Who Should Handle What?
Installation & Setup: Agent, it was fast and runs in background
Overall design & layout: Agent, the default design was decent, but my design-first approach might’ve saved fine-tuning time
Front-end components: Agent was fast
Debugging the API: Me, Agent took longer and cost more
UI fine-tuning: Me, Agent followed prompts but I’m still faster and free
Publishing: Agent, super convenient
Code quality: Me, but Agent was fine
Thanks for reading!
If you found this article useful, please like and share it with someone who might enjoy it.
And let me know, have you built anything with Replit Agent? Drop a comment below, I’d love to hear about your experiences.
Top comments (0)