Replit

Our verdict on Replit

Replit was Jordan's default in 2025 for small internal tools that need a database, a backend and stored API keys. Six of our builds ran on it, including the AI video studio from our first episode. Sam called it probably his least favorite vibe-coding tool, and by April 2026 Jordan used it mainly for apps he plans to throw away.

What we built with Replit

• Ryz Labs Video Studio (June 2025): an internal app where the team picks video models like Google Veo 3 and Seedance and image models like FLUX Pro, enhances the prompt with AI and shares a history. The first version shipped in a few days. A video editor made 20 to 30 clips in about an hour and had a commercial within a day. Some image models still failed, or images disappeared from storage.
• Ryz Labs Autoblogger (July 2025): built mostly in Replit, it writes illustrated posts for each brand and publishes them to Webflow. About 150 posts per brand since launch.
• Podcast CRM (July 2025): a list of the top 1,061 business podcasts with contact emails from their RSS feeds, plus AI-written outreach that opens in Gmail. Replit grabbed and sourced the podcast data for him.
• Public API trading bot (August 2025): front end and back end built in about three to six hours. ChatGPT decides whether to buy or sell the IBIT ETF every 30 seconds, with prices from Polygon.io. It turned $300 into $299 over about ten to twelve trades. As Jordan said on air, this is not investment advice.
• Ryz Chef (September 2025): upload a grocery receipt and get 11 Mediterranean recipes. AI-generated meal photos were too slow and expensive, so Jordan switched to Unsplash images.
• Trick-or-treat AI costume builder (October 2025): a webcam photo goes to Nano Banana, then to Veo 3.1 Fast for a Halloween video. Heavily used inside the company in its first 48 hours.

How we use Replit

1. Pick Replit only when the app needs a backend. Jordan's rule from Episode 1 (at 20:04): Replit for anything that needs an API, database or auth; Bolt or Lovable for front-end demos.
2. Improve the first prompt outside Replit. For the video studio, Jordan had ChatGPT sharpen a one-line idea: a web admin panel with a left-hand sidebar that generates video with different models. For the trading bot he probably had ChatGPT write a PRD.
3. Use the built-in services. Database, storage and environment keys are already there, so API keys for OpenAI, Claude or Replicate go straight into Replit.
4. Add models by pasting their API docs. On Replicate's API page, copy the sample call and the JSON schema, then tell the Replit agent to add the model.
5. Let the agent fetch data and know APIs. It sourced the podcast list itself and already knew most of Public's API; Jordan pasted docs only when needed.
6. Try the planning feature. Sam pointed to it in Episode 18 as a way to make builds easier and sturdier before building starts. We did not show it on air.
7. Iterate in chat, then redeploy. Each new feature is a chat message and a click on redeploy.

But when I wanna build something that I need like an API and I need to make some some functional calls, I'll go to Replit.

— Jordan Metzner, Episode 1

Where Replit falls short

• Scale. Even Jordan frames it as a tool for things that will not scale to a huge organization. For the trading bot's next version he wanted a Supabase backend, a React front end and better cron jobs.
• Not everyone likes it. Sam's take in Episode 12: probably his least favorite, which shows there is a vibe-coding tool for everyone.
• Unfinished pages. Parts of the trading bot, such as AI predictions, did not fully work, and the video studio lost some images from storage.
• Big software goes elsewhere. In Episode 38 (at 15:47) Jordan put big builds in the Cursor, Claude Code and OpenAI Codex vertical, and Replit with Lovable for quick throwaway apps.

If I'm building something that I need a quick database and I need it to live for a little bit, but then I'm gonna throw it away, Then it kinda falls into that, like, you know, kinda like Replet lovable ish world.

— Jordan Metzner, Episode 38

We did not discuss Replit's pricing on the show; the only costs we covered were the model APIs our apps call.

Replit compared

• Bolt vs Replit: Replit when you need an API, backend, database or auth; Bolt for front-end-only product demos.

Episodes featuring Replit

• Episode 1: the video studio, and Jordan's rule for when to use Replit.
• Episode 2: the Autoblogger.
• Episode 3: the Podcast CRM.
• Episode 8: the Public API trading bot, built just by talking to Replit.
• Episode 12: Ryz Chef, Replit's funding round and Sam's dissent.
• Episode 18: the costume builder and the planning feature.
• Episode 38: where Replit fits in Jordan's tool framework now.

FAQ

What is Replit good for?

Small tools that are not front end only: they need a database, a backend, some auth or API calls, but will not scale to a huge organization. Jordan built our AI video studio, autoblogger, podcast CRM and a trading bot this way.

How do you use Replit?

Write a PRD or improved prompt with ChatGPT, give it to the Replit agent, store your API keys in Replit, then add features by chatting and hit redeploy. For new AI models, Jordan pastes the sample API call and schema from Replicate and asks the agent to add them.

Is Replit better than Bolt?

For different jobs. Jordan uses Replit when an app needs an API and a backend, and Bolt or Lovable for front-end product demos.

Do you still use Replit?

Less. Our Replit builds ran from June to October 2025. By April 2026 Jordan put Replit in the bucket for apps that need a quick database and will be thrown away, with big software going to Cursor, Claude Code or Codex.