Replit Agent 3 is here
byβ’
Agent 3, Replit's most autonomous agent yet π₯
Check the launch here: https://www.producthunt.com/products/replit?launch=replit-agent-3
What's new with Agent 3:
Tests & fixes the App it builds automatically using a browser- checking buttons, forms, links and APIs.
Runs for 200+ mins autonomously as it is building, testing and fixing your app- eliminating the need for manual oversight.
Builds Agents and automations to automate complex & repetitive tasks. You can now build Slack or Telegram bots or workflow automations integrated with Notion, Linear, Sharepoint, Dropbox & more in plain English.
Share your thoughts here. π
78 views


Replies
I have been using the new Agent 3 for a couple of days now and here are my findings...
1. AI takes a little bit longer to process your prompt and it appears it has been slowed down a bit. This is all good though because it is taking the time to understand, code, test, fix, etc.
2. The AI has the help of other AI to help with its tasks and to validate that it's reasoning is correct or not.
3. AI has the help also of an Architect AI to help with proper ways of coding, testing, UI/UX architecture, database architecture, & Security just to name a few.
Does it get everything right from the first prompt? Absolutely not, but I do appreciate that it is taking it time, checking itself and constantly checking that my app, database, and security... Basically the whole architecture is being checked and double checked along the way.
It still struggles, but I think a little bit less and I believe these changes are right on track for what was needed. Of course this is just one person's opinion and you are welcome to yours. For me, it's all good!
My experience was absolutely Great - before the introduction of this Agent 3 - I had launched 4 Products here on Product hunt. All developed with Replit - And I repeat, I was HAPPY with Replit - Gave it 5 stars!
But now, sadly, this is how my experience (Hope this is just problem with my instance/environment probably) has been since the Agent 3 launch,
Here is a summary of the user's complaints in bullet points:
Frequent Technical Issues: I'm experiencing a recurring Vite dev server caching issue, which leads to wasted computing resources, financial losses, and causes the AI Agent to perform incorrectly ("hallucinate").
Degraded AI Performance: Recent updates have made the AI Agent "delusional" and less effective than previous versions, forcing the user to repeat prompts in frustration.
Poor User Experience (UX): The platform has become less user-friendly for non-technical users. Specifically, the app deployment process, once a simple one-click operation, is now a troublesome, multi-step process.
Unsatisfactory User Interface (UI): I'm finding the application's theme to be outdated ("old rustic"). I have pointed out specific UI bugs like overlapping menus and poorly designed dialog boxes. and eventhough the Agent says it fixed it. but in reality it did not.
Failed Fixes: Despite the AI's attempts to redesign the UI with a modern, colorful theme, the final result is "Completely Unsatisfactory".
Overall Dissatisfaction: Due to these persistent issues, I'm feeling frustrated and considering finding an alternative service unless significant improvements are made.
Earlier it was on auto-pilot for most part of the app development and deployment. But now I feel it needs more spoon feeding, as it keeps forgetting, skips steps.
"Sincerely Hoping" improvements are made.
Cal ID
Replit is truly amazing. It does so much without needing much user input. Hoping to get some updates on back-end as well. I still think it needs some refining
An AI agent that actually critiques its own work is quite powerful. Good stuff! I'm enjoying building with this new Agent 3.