This is a popular fallacy, exposed many times, most recently by Debbie
Lapeyre in her closing keynote at the MarkupUK Conference in London a
few weeks ago.

1. As Michael Kay rightly pointed out, it is probably true at a small
   scale (as are many claims to simplicity, not just in IT).

2. JSON is a good fit to shallow models of rectangular data (rows
   and columns) particularly where all components are present in every
   "record", even if sometimes set to null.

3. Because of its syntax, it can be absorbed at high speed with little
   excise compared with XML for similar scales and models.

4. JSON cannot handle mixed content at any meaningful scale, so it is a
   poor fit for deep data with complex conditions for optionality.

> My personal feeling is, that people would choose XML when they're
> developing a big software instead of a tiny one. And when good
> software engineering is important.

If this is true, then the software industry is in a worse state than we
imagined :-)

Developers and programmers appear to have a tendency to instinctively
choose JSON possibly because they believe it is a data format suitable
for their programs to read. Perhaps because they tend not to have much
experience with markup languages, and possibly because they have been
seriously misinformed in the past about how to use them, they tend to
steer away from XML.

There are many factors which influence the choice of file format, and it
really isn't possible to make very many rules about which one to choose.
As Ihe Onwuka said, JSON may be regarded as simpler because it has so
far been used for simpler things. Debbie's point was that it cannot be
used for complex TEXT documents because it cannot meaningfully model
mixed content. I don't think it is necessarily related to project size
at all, except that bigger projects may have people with greater
experience working for them.