Build the UndrVibe app mostly with an agent
The poster built their first real app, UndrVibe, mostly using Replit Agent, and shares the cost of agent versus assistant usage.
Done withReplit Agent
- What they did
- The poster says they built UndrVibe in Replit mostly with Replit Agent, using the highest model/settings with testing turned off. Agent handled most of the build and also cleaned up its own messes, while the poster switched to Assistant for one-off single-page edits and troubleshooting when features broke.
- How it went
- By that point, they report spending about $362 on Agent versus $4.50 on Assistant across 90 prompts/edits. The takeaway was that Agent did most of the heavy lifting, but Assistant was much cheaper for targeted fixes.
- Worth knowing
- Their clearest caution is cost: even on one first real app, Agent spend was far higher than Assistant, and they specifically mention turning off testing.
Try it yourself with Replit Agent
I'm building my first real app: [app name and one-line description]. Start with the core feature [describe it], build it end to end, and explain what you did in plain language. Keep changes small and tell me roughly what each step will cost in usage before big tasks. You're done when I can use the core feature in the deployed app.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · Undated
Related use cases
Build a searchable map of funded companies
A software engineer had Replit Agent build a searchable map of Wefunder brick-and-mortar companies in about 3 hours for about $22.
Build a Pokemon-style dog collecting app
A user had Replit Agent build a catch-em-all dog app whose rarity comes from scraped American Kennel Club regional population data.
Publish four Android apps to Google Play built with Replit
A user built four apps with Replit and published them to Google Play, with Replit writing the GitHub workflow YAML files for builds.
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Run sandboxed Claude Code fleet sharing files via MCP
A developer runs a fleet of sandboxed Claude Code instances that share files through MCP tools on a virtual filesystem, with an orchestrator handling AWS storage and audit logs.
Build a local knowledge app for agent outputs with Codex
The author built Mianotes using four Codex agents across 30+ sessions to centralize scattered files agents produced, and the agents documented the product inside it.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.