First, let me say i’m not bashing Bolt. I’m just trying to do some information gathering. I’ve used both bolt.new and bolt.diy a couple of months ago and to me, I struggled enough with both of them that I didn’t feel either were production ready. And i’m totally open to the fact that it could just be the way i’m using it, which is why i’m asking here, if it is a production ready app.
I would love to be able to use the app on a daily basis.
I would also love to see if there are any examples of complex apps that people have built with it. I could build very simple games, but it seems like the more complex the app by adding authentication and admit our security section, the more it just does not work the way I intended to.
The app is in very early development state and just starting up with advanced features. Also the version number (0.0.5) tells you that we, in my view, far away from production ready and reliable.
So I totally agree which your statement, that it is not capable of doing complex projects with databases and authentication just out of the box (maybe with a lot of workarounds you get it to work).
But it will need at least several months until we hit a state where you could work “normal” with it.
Thanks, @leex279 . This is great information, and I appreciate the transparency. I definitely want to play with it more and use it. I am fascinated with these apps and love that I can run them locally.
I would argue that ALL AI development tools currently feel unpolished and more like beta software. Including AnythingLLM, Cline, Windsurf, Back4App, Devin, Bolt.new, OpenHands/Daytona, Copilot, etc. And there are tons of projects trying to compete in this space and many of them are commercial. Imo, they need a lot of refinement, optimization, and improvements before they will become truly useful.
From my testing using many of them, I find AI frustrating to use on complicated projects… and the stupidest things improve the results… often system prompting, breaking up tasks into “steps” and so on. And then of course there are ways to make the AI “reason”, such as RAG, Tool Use, giving longer “thinking” time, etc.
One thing that all these tools seem to have a problem with is memory bloat and eating through tokens like crazy. And I personally think an n8n workflow can solve this almost entirely. That’s what I’m personally going to work on for the Hackathon, but I just decided today!
Basically, it feels like we are being forced to learn the AI rather than it learning from us. It seems things are improving daily though, so I’m still optimistic that AI will get more and more useful, with less frustration and pain points. But who knows? Lol
