I and some other people are working on ways around that limitation.
How?
And I'll tell you what I am currently doing. I have a Mac Mini as a home server, running an app called HAM Bridge. SmartThings events send an HTTP GET with various params to HAM Bridge, which then hands them off to an AppleScript for speaking on the Echo via bluetooth.
The holy grail is to, if possible, develop an app on a RPi (or maybe even an Arduino) that acts as a bridge between the audio-over-WiFi that ST supports (driving stuff like Sonos, et al) and the Echo's remote control interface (which is Bluetooth) and the "Simon says..." function to send speech to it, which the Echo then converts to text and then speaks in its own voice.
But in the meantime I already have something almost as good. I'm using a combination of Rule Machine and the Lannouncer device on ST to send notifications to the Lannouncer app on my Android phone (until I can setup a more dedicated Android device for that, like a cheap old tablet or something), which then converts it to audio and plays it. Since my phone is paired with my Echos the audio is then sent to the Echo via the regular Bluetooth interface.
Right now I only have a couple of a proof-of-concept rules set up that tell us "Good night" when the bedroom lights are turned off for the evening, and tells me, "Come on old man, it's time to get up" right before my alarm goes off in the morning. I have additional time of day conditions on the "Good Night" rule so that it only fires between 10pm-1am, and I use similar conditional rules on my phone's Tasker app to only connect to the Echos' BT at that time, and disconnect in the morning after I get my wake up nudge.
It's a little kludgy, but it works.