I wrote a large request to create a website with 10 pages and described each of them. The AI started working. It made main.ejs quickly enough. And then he did index.ejs for 24 hours.
The AI was writing all the time: "I need to provide the required parameters for the set_current_page_content function. Let me create the complete homepage with all the required sections."
sometimes it's something else, but he wrote this phrase exactly more than 100 times.
As a result, he generated index.ejs. Now he has started generating the About page He made this page much faster - in about 20 minutes.
Well, of course it's good that it's working, and I'm resting, but I think something's wrong with the algorithm. The idea of AI is just great, and I'm sure it's fully feasible, and I'm really looking forward to it working fully. Now I use AI a lot in my work and I miss it in Wappler
Agree with you. Moreover Claude 4.5 (especially Claude Opus 4.5) has full knowledge of Wappler framework. I tried Opus 4.5 (as experiment) and it worked very well modifying a wappler website, doing serverconnects etc (not 100% but 95%)
Thank you and Cheese. I'm currently editing my prompt. I'll try using ChatGPT and Claude. I'll compare 3 work options and write my result here, well, and show you what happened in the end I want to make a website entirely using AI (at least the first version).
I uploaded a fairly large (22 Kb) query to the AI Manager (Claude Opus 4.5). The window for tracking his actions has been reduced to 1 line. You can see it in the photo. I can't zoom in, it's not convenient.
The first attempt resulted in a stop with the message "Overload". I ran the same query again and the generation started.
As a result, I got the error again. Wappler created only main.ejs and started making style.css. And it cost 1.5$. The entire website on DeepSeek cost me $0.5. So far, it turns out that using AI is quite problematic. I will try ChatGPT and if there are such problems, I will try to write requests to each page of the site.
Last error:
429 This request would exceed the rate limit for your organization () of 30,000 input tokens per minute. For details, refer to: Rate limits - Claude Docs. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens requested, or try again later. You may also contact sales at Contact sales | Claude to discuss your options for a rate limit increase.
Switched to gpt-5.2. Work began, and the main.ejs page was generated fairly quickly. At the same time, its appearance is the best of what other neural networks have produced, including Claude Opus 4.5….
I observe the following behavior. Wappler AI created the page. After that, it goes into some kind of cycle of reading the code and applying the changes. However, no changes are visible. The page remains the same, but the cycle continues. This not only wastes time, but also money, as Wappler constantly sends code to chat-gpt. I have observed similar behavior when using DeepSeek. This did not happen with Claude, probably because he immediately went into error due to the limit of tokens per minute.
I was trying to ask a query in which I described the entire website. I see that when creating pages, the subquery for the page (which Wappler does itself) may be very different from what I requested. In general, I admit that the experiment with a query describing the entire website was not very successful. I will try to make each page separately.
Once again, I note that in the AI Manager, the window with information about how AI works is very small and nothing is visible. Scrolling works, and you can see that there is a lot of information, but it is very inconvenient in this format.
Wappler makes new site pages again with the same behavior. He quickly makes a page, and then enters a long cycle of some changes, but these changes are not visible. The page remains exactly the same. After 5 minutes of such strange work, he finishes and moves on to the next page.
To view the changes in the "endless" loop, I switched to the code. And here is an example of what Wappler does many many times. In this case, it changes one line of code to exactly the same one. Another time, he does it with a different section of the code. That's why I don't see any changes visually. Why is he doing this? In my opinion, Wappler makes 90% of unnecessary requests to the neural network. As a result, the API fee can be reduced by 10 times or even more.
Wappler AI created all the pages, in general they didn't look bad. Next, he created style.css and everything became black and unreadable. And then he started redoing ALL the pages and started changing some of them completely. In one case, this led to an error and instead of the finished page, I received a text on the page saying that something went wrong and needed to be fixed.
As a result, Wappler used up all the $20 that I paid to ChatGPT. At the same time, for a 12-page site, he made 612 requests to the ChatGPT API (5.2) and sent him 62M tokens. Watching the work, I have no doubt that >90% of requests were made by mistake. In the end, the work was completely ruined by the fact that he started to redo everything. I couldn't check when it would have ended because the money ran out. So far, I conclude that it is impossible to write extensive assignments to him. I will try to work in a different way.
However, it would be great if the window with information about the work of AI were larger (I wrote about this above).
Of course, you also need to get rid of multiple duplicate requests and "changes" immediately after the page is generated.
And at the end of the job, Wappler doesn't have to start it all over again.
I hope my comments will be helpful. As I said, it's a great idea, and I'm sure it can be improved. If necessary, I can send my prompt and you can test it yourself.
I use AI for JavaScript coding but the few times i tried Wappler coding, the resulting code took longer to debug than it would have taken to write.
My one big fear is this forum becomes dominated with posts asking those who have learned to code in Wappler to fix AI code produced by those who opted to take the perceived easy route.
I use AI when programming in python, it also helps well with ansible and more general things. But I always control what the AI does, I don't need code that I don't understand. In the case of Wappler, AI actually does the markup, not the code. This is what I saw, AI writes html markup. And this is a very correct task for a neural network. There's not much to debug here. I haven't thought about writing code in wappler yet. When the Wappler team announced the introduction of AI, I was internally against it because I already don't fully understand how Wappler works, and if AI does something else, then control will be completely lost. I believed and still believe that it is better to spend time improving the wappler itself than chasing fashion. But since AI was implemented, I decided to use it. And I think that AI can be very useful for Wappler in layout tasks, and once the introduction of AI has begun, it can be completed.
