Quite obviously more and more Users are turning towards AI for solutions. The sad flip-side of that is we are losing points of discussion and the thrill of the chase for those of us that like to help. We are also missing the resolutions to the issues that arose, the route to the resolution, and with that helpful context for other Users who may be experiencing the same and need guidance.
Thus I'd like to ask what issues you have solved and how you went about it?
What was the issue?
What model did you find most helpful?
What was the resolution?
What prompts have you found work well for Wappler?
And if you are stuck in a loop and haven't got an end-result yet then please feel free to share it with us all.
If AI is going to draw Users away from posting their issues then lets talk about how you use AI in your workflow and how you resolved them. It could really help others in a similar position.
For me I quite like the Wappler ai using Claude. I use it to make quite a lot of my front end stuff. I really don’t trust it though to do anything with server connect. I found it creates steps you can’t see in the tree view so you are clueless what’s happening.
GitHub copilot limits are now really bad though. Thinking of trying open router for wapplers ai.
My latest use of Ai was to get mqtt working front end using Gemini and then using this to update a data store. Really do like Gemini at the moment. Also getting Geoman on leaflet maps again with Gemini. Both took a couple of attempts but got there in the end.
Hoping next to get it to help me to setup timescale on Postgres.
I find it doesn’t figure it out in the first 3 tries I am better reverting my changes and amending me prompt.
I very seldom use Wappler AI. Certainly have never used Act mode. I find the responses way too long in length and impossible to read it all. About all I have accomplished with Wappler AI is getting some CSS advice. Even then the response is about 10 screen lengths long when I just need the simple CSS Rule.
I find for my use just using external ChatGPT is way faster and to the point. I personally don't see Wappler AI becoming a goto source in my workflow. This of course could change.
I'll be honest we have not used Wappler AI since its very beginnings. No reason behind that though. Our work-flow was already established prior to its implementation so I can't really comment on it. We are huge fans of Open Weight models. No particular order to our preference:
Qwen family - built some astonishing internal applications for a couple of Euro's.
Deepseek family - superb family of models.
Kimi - another fine family of models.
Meta Llama 4 (amazing image recognition).
We were fans of the Claude family but in the past few months there has been some sort of shift and one of our favorites, Sonnet, just become useless essentially. Such a huge token waster. As did Opus. So we avoid both now. As for Open AI never really been a fan so bias reigns supreme in that regard.
For the most part we try to use the above open weights for a lot of work. We also chain them together so they argue among themselves which I've mentioned in other threads. We'll run them in parallel which seems to get the best from them.
With regards to prompting we try to build a knowledge base based on our thousands of Actions from dozens of Projects and separate those in to groups such as security, insert, update, delete, image handling, S3/storage, mailer, etc. We then amalgamate the findings in to single files which we then pass to the models themselves, and ask them to review them and subsequent documents they refer to. We also log all discoveries and idiosyncrasies, structures, and errors. On top we fuzz a lot of things to cause edge case errors and learn how to handle those before we accidentally introduce them. Always a strong focus on debugging. Multiple versions then run smoke tests for comparison with regards to resource needs. Then we loop again and again and again. For the low cost it doesn't really break the bank so no harm there. Result is clean and performance focused output which we can implement in to production. We are always learning new things.
A bit like cooking really. We are all sharing the same ingredients for the most part. But our dishes are all slightly different. Suppose that is like music too. Only so many notes!
