Replit Agent and pre-existing code

I’m building a social app on Replit. I’ve previously built a version of my app entirely with Replit Agent but I had problems with nested replies, sharing posts and many other small features that would be essential for a seamless user experience. Now I’m trying to just clone the frontend from Bluesky as it is open source but I’m not sure if the Replit Agent is good at working with a pre-existing codebase, especially one so large. Does anyone else have experience with this?

@realfunnyeric may be able to help here

I personally wouldn’t bother with Bluesky and would build everything to spec in Design mode, exactly how I want it to work, then convert to app and go to town making it work. It’s perfectly capable of doing so.

1 Like

So don’t clone an already existing codebase? My main issue last time was that there are many social media features that the Replit Agent just couldn’t get right no matter how hard I tried. Things like threaded replies, the homepage feed and sharing. I created the bulk of it around early summer this year.

Again, just my personal opinion. I’d like to build it.

Threaded replies and activity feeds are where AI agents can hit their limits. The logic gets messy fast with nested comments, real-time updates, and notification patterns. I’ve seen devs burn weeks trying to get these right through prompting.

Drop-in components handle this stuff in minutes. Weavy’s Feeds component does threaded discussions, reactions out of the box. Messenger handles the chat side with proper threading and file attachments. Lets you spend AI tokens on your unique features instead of fighting with WebSocket protocols.

2 Likes

I built one that worked pretty well back on Agent 1, and I can only imagine it’s gotten better now as models have advanced. Drop-in components can carry their own baggage.

Lots of ways to skin a cat! Pick your poison.

Cloning Bluesky’s frontend is ambitious but you’ll hit the same nested replies and sharing problems. The complexity isn’t in the UI patterns, it’s in the real-time infrastructure behind them.

You already know from your first attempt that prompting agents to build social features burns tokens and still leaves gaps. Drop-in components for feeds, messaging, and posts handle that infrastructure in about 10 minutes. Weavy’s designed for exactly this - social features that work without re-prompting the agent every time you need threading or reactions.

What specifically broke with nested replies in your first version?

1 Like

Hard to describe exactly but I think the most obvious issue was the failure to even show all replies or posts that were several replies deep. Not being able to ever view an actual thread may have made other issues even harder to discern on top of this.