I’d also like the models to stop charging me tokens for the same errors repeated over and over again but I think that’s Claude not Q. Need to dig into that more. Telling the model in context not to repeat the same mistakes doesn’t work either.

Pay as you go pricing for Q CLI like every other service as is the original premise for AWS services with a way to cap spending for those that need it.

Don’t force me to use IC. It’s very confusing and overkill. Just tried to set up new user in IC again to use Q App and was royal pain. Wasted a lot of time.

What Q wanted to do is probably an ok practice in rust but not really what I wanted. Maybe I’ll revisit it later but I’ll see how it works out. I just never want certain libraries to crash. We’ll see.

My latest attempt to address that is a structure that forces the agents (as much as they can be forced) into writing proper error handling with meaningful messages that explain exactly what is wrong and how to fix it. That is my latest experiment.

Speaking of errors, that is what I am working on. Consistent error handling. I’ve written about how important it is but even more so with AI that keeps making repeated mistakes and is swallowing errors, making things difficult to troubleshoot.

I could rush and vibe code some tool to do what I’m trying to do, but I am hopeful that building a solid base platform for AI development will speed things up and keep my code more consistent with less wonky errors.

I had these aha moments while building it where I could wipe out mountains of code and simplify everything that were hugely satisfying. I’m aware of the pitfalls and how to leverage minimalist code to do a lot.

I learned a lot of the pitfalls of building a development platform and architecting software (by doing it incorrectly at times and creating circular dependencies that were impossible to maintain and then fixing it.)

I still don’t really have the energy to deal with all that again. That said, what I learned building that platform has helped me immensely as I make my way into the AI space.

I think I was also a bit burned by what I saw during the dot bomb and how one of my clients treated me. Paying my employees’ healthcare out of my own saving was also a bit rough.