Here’s the reality - v0 is an awesome tool, and can build some great stuff, but it produces errors and the wrong result most of the time. Yesterday, I wanted to get it to do a simple thing (adding an email gate to accept a verification code) - i didn’t even want it functional, just producing the experience.
This took about 15 prompts, before I just gave up - and after reviewing I blew through about half my monthly budget.
I understand this tool chews up a ton of tokens, but unfortunately, it is very uncommon for v0 to deliver what you want on the first shot. It usually takes trial and error, which is fine with a flat monthly fee - but when you are being charged for errors, it becomes a broken experience.
I have also found that you basically need to yell at v0 to not make changes. Sometimes I want v0 to suggest a change, rather than execute one and this doesnt seem to be built into the model.
The product team could easily make this token based system a clearer ROI for their users through a few builds:
-charging people ONLY when they accept a change. If a change throws an error, or is wrong, it should revert back the previous session and not save. This way, people pay for builds they like, and I would assume v0 will get data back to improve their models.
-creating a feedback loop prior to executing a change to the code base. Meaning, the AI can see your request, suggest a change - and only implement if IF the user agrees it’s right. This will dramatically save output tokens.
-suggesting which model to use. I see v0 implemented a medium and large model - and in my brief experience, it was not clear if either was better. I defaulted to the large one after frustration yesterday. I would imagine their systems could be smart enough to determine which model is most appropriate for either change - and again, confirm that change with the user.
-creating a better way to track credits/usage. There is simply no way an end user could see how many output tokens a change will take, so expecting a user to have a change of anywhere from a few cents to a dollar for one execution is a broken experience.
In one day, we’re already discussing new tools internally. Its a great product, and we may still use it for some prototyping, but to make such a dramatic change like this without letting users get a decent sense of what was coming is something I haven’t seen in all my years subscribing to SaaS products.
