I tested Agent 3 on Replit, and honestly, it was a very frustrating experience.
In my opinion, it is currently the weakest agent. It often ignores instructions, works independently in a different direction, and produces errors instead of completing tasks correctly. What’s especially frustrating is that after fixing one problem, several new issues often appear.
It gives the impression that the system is more focused on maximizing usage rather than functioning reliably. The agent overcomplicates things, rebuilds entire structures unnecessarily, and in the end implements its own version of the project instead of following the given instructions.
At the same time, as a user, it feels like you are paying for something that still lacks the necessary stability and quality.
By contrast, I use Verdent and must say it is the best program for me. It completes everything fully and prepares projects for any platform. The results are clean, organized, and work without errors. This is exactly how a functional system should operate.
I hope the Replit developers take this feedback seriously and work on reliability, accuracy, and better adherence to instructions.
Best regards
