However - slightly disappointed that the programs are not literally controlled by symbols on the pieces of paper, but rather the paper just holds identifiers who point to a program on a server.
I wonder if you could extend the computer vision system to pick up actual programs from the paper - either by OCRing actual text (perhaps limited to capital letters only, and/or particular shapes for letters such as I-for-India, to make it easier?) or a combination of symbols (each corresponding to a function in the API) and numbers (parameters).
That way you could do really interesting things: modifying somebody's program by literally pasting a scrap of paper on it containing different symbols or parameters, or having programs which interact with other programs (either by treating their program code as input data, or by piping data between the programs).
On the other hand, it's probably easier to write interesting programs with an actual keyboard :)