Then there's the (slightly oldish) Pictures page on the wiki:
http://wiki.kogics.net/sf:pictures

There's also some material on them here:
http://lalitpant.blogspot.in/2012/05/recursive-drawing-with-kojo.html

First of all, I wish there was a way to run Boxer, to see the ideas in action!

I'm personally not a big fan of visual programming (I think it makes the high-ceiling of an environment lower, and the wide-walls narrower). But there's an argument to be made that it provides a lower floor.

So how can these ideas make Kojo better?

We need to make the workings of programs more visible to users. An important effort in this direction is the script-tracing GSOC project that's being done this summer.

We need better syntax-error recovery (these don't happen at all in a Visual Programming Environment).

I was also struck by the similarity, in a lot of respects, between Boxes in Boxer and Pictures in Kojo. It might be an interesting idea to slap on a nice UI over pictures to see where that leads.

In any case, thanks for the very interesting links!

(Testing if I can post links.)

It doesn't look like this has come up on the forums before, and there
are a lot of good ideas:

http://www.soe.berkeley.edu/boxer/20reasons.pdf
via http://lambda-the-ultimate.org/node/4695

I put a link to kojo in a comment on Phil Wadler's blog where he has been discussing languages for kids

Great!

Needless to say kojo doesn't have a low ceiling.

Exactly right. Kojo's got a low floor and a high ceiling (courtesy Scala).

From your comment:

My 7-year old grandson loves it, but it does need a bit of adult expertise to explain error messages at the moment. The author is working on it.

I'm afraid this is on hold right now due to time/prioritization constraints.
After I added Stories to Kojo, the subset of Scala used by kids became much bigger (and the task of defining language levels for Kojo became much more complicated).
Also, the kids that I work with seem to get by just fine when they encounter errors. Most of them don't pay much attention to the error messages, but, given that they know (with Kojo's help) the exact line number where the error is, they are able to examine the location of the error - and (mostly) figure out and recover from the problem.