Sessions discusses some of the key problems in enterprise architecture and identifies complexity as the primary obstruction to resolving these. He uses some fairly simple maths from probability & set theory to both characterise the problems and as basis for his proposed method of tackling the issue - iterative partitioning of the domain into "autonomous business constructs" and strictly defining the boundaries and interfaces between them.
As one of his prime examples, he dissects the NHS' National Programme for IT, identifies the inherent complexity at its heart, and predicts its total failure. Which admittedly, could be said of pretty much every IT professional in Britain who was not working on the project, but his explanations of exactly why the project's attempted risk mitigations actually made matters worse, while increasing complexity at the same time, are well worth reading (and should be mandatory for anyone commissioning large government IT projects)
Throughout, he make his case well, and his methodology is likely to lead to less complexity in the enterprise architecture, but reading the latter half of the book, I couldn't help but feel that there's not actually a lot new here - for "simple iterative partitioning" and ABCs, read domain-driven design and a refined class model, albeit taken to a somewhat higher level of abstraction.