As a reminder, I wanted to use AI to see how far I could go in creating a concept chatbot. During November (which in AI land is generations ago), I dedicated significant time to exploring Replit, an IDE designed for non-coders. This powerful tool allows users to prompt and build projects, taking them all the way to deployment in the real world.
Throughout this process, I engaged in over 114 checkpoints with the coding agent. This involved communicating with the agent to request edits, overhauls, or new builds. Depending on the complexity of the request, responses could take anywhere from a moment to over 30 minutes.
These checkpoints accumulate because, while the agent is often successful, it occasionally rebuilds entire sections of the application, forgets existing code, or neglects to include essential components like the frontend UI. For instance, I had the agent create an integration with Intellistack, which was an impressive feat. However, when I requested an edit, it ended up coding a completely new version of the integration, which functioned differently than intended.
At times, the agent may freeze or become overwhelmed with context, resulting in lengthy periods of inactivity. So far, I have invested approximately $200 in this project, a fraction of what it would cost if a person were handling the work, which could easily reach the thousands or ten-thousands.
What’s amazing to me is that at the end of less than 10 days of back and forth with Replit, I had a fully functioning tool that met all of my original requirements and was hosted on my own domain.
That felt powerful!
Here are a handful of screenshots of the working product.
Chatbot Initial View (with Streamline webhooks)
With all of that energy for what I’d built…I knew that something was still amiss. I knew that I didn’t know a lot about what I’d built. Was this application truly ready for prime time? I needed some real life help.
