An interesting dynamic here is that, having read Yeggie's article years back, I regarded Smalltalk as being somewhere towards the conservative end of the axis (at least, in comparison to other dynamic languages.) E.g.: variables declared before use, no truthiness, very few automatic type conversions. It results in a system with a high degree of internal consistency.

And yet, the goal of the language is to enable activity at the liberal end of the axis.

So maybe there's a more subtle goal: you have to enable someone to make a mess, but also equip them to get out of that mess into something more structured, as and when that becomes useful.

As a counter example, spreadsheets are wonderful for making a messy solution to certain kinds of problems - but how do you then it them into a spreadsheet that isn't a mess? I don't know that I've ever seen that happen - it seems like the right means of abstraction and combination to enable that to happen just don't exist in spreadsheet-land.