spoiler alert: auto scaling is far to be magical

as usual with aws, everything is an object and for autoscaling u got several objects and several links between them
First you define a launch configuration : what type of machine you want to laucnh
Then an autoscaling group : it will use the launch configuration to create new EC2
In this autoscaling group, you have to define auto scaling policies ie what to do (remove or add x instances) and link them to a cloudwatch alert (cpu is high or network or whatever cloudwatch monitors)
the autoscaling group can also be linked to an ELB so when EC2 are added/removed, they also are registred/deregistered from ELB

A lot of work to do..

Example : I change the autoscaling trigger in web UI interface. After a few second, my autoscaling trigger is changed (checked).
But if I save beanstalk environment configuration into a file, the change I made is not there :-(

Maybe working wih UI is not the best option, but it's very convenient to test stuff without terminate/recreate env from scratch (things I have to do btw when I commit changes in files, I have to check if the behaviour is the same)

A lot of time lost, but it's the price to have reproductible environment and 'infrastructure as a code'