Yes, dictation is completely different. And also completely different from the subject at hand. When you were dictating, the computer didn't have to perform tasks. It just had to output text.
True, sending a Zigbee cluster is quite taxing. I tried doing it once and went horse halfway through and had to stop and mix a drink. Mix? Drink? Awesome idea, I don't mind if I do!
Just a note or two about Mycroft - I have used it for a captive project. A Pi 3 is not fast enough. A Pi 4 4GB is good enough. The inventory of skills is slim. Recognizing words is not speech recognition and Mycroft is just so-so on either so the total is less than the sum. It has lots of integration possibilities if you know Python and a lot easier to program than C++ for the Alexa SDK. If you've looked at the Alexa SDK and eco-system you'll know just how far mycroft has to go. It's also amazing that an open source project has gotten that far.
Just wondering, what makes you say Alexa requires c++? You can use just about any language including python. Alexa Skills Kit SDK for Python | Alexa Skills Kit
Back when I was looking at it, only C++ had the subsystems/capabilities I wanted to use. Python only did about 3/4's of the full API. It may well have improved since then because once I decided mycroft would work for me that's all I looked at. I do remember being happy not having to work in C++ on those libraries. That's another reason I never re-evaluated Alexa SDK, Too much work for a hobbyist. If it was a 'real' project/product, Alexa would have been best.