In "The Self-Organizing Economy", Krugman explained why he thought that economic geography had died out sometime in the 1960's. Partly, he said, it was that the discipline lacked "microfoundations": it didn't explain high-level behaviors (in this case the existence of cities) from the unguided actions of individual economic actors. Instead it took the existence of cities as given, then derived conclusions about where people and businesses would locate. "The Self-Organizing Economy" painted some cute little models to try to build these microfoundations. Widely dispersed populations turned out in that book to be an unstable equilibrium: we get the microfoundations by assuming a "state of nature" in which everyone is spread out, then show that the state doesn't last. Krugman actually comes to a stronger conclusion from his toy model: cities end up being evenly spaced around the circular landscape. Any closer together and they start eating into each other's markets. Any further apart and they lose the benefits of closeness to customers and suppliers. This unifies a number of traditions in economics that have tried, over the years, to explain why cities exist in the shapes and sizes they do.
"Development" assumes more economic knowledge than did "The Self-Organized Economy", though I could fumble along and get most of what he was saying. Understanding why cities concentrate at all, says Krugman, inevitably means understanding increasing returns to scale. My intuition is ill-formed here at the moment, but I think the idea is that with constant returns to scale, doubling the number of employees in a given factory only doubles your output - so there's no reason to prefer one large factory to two small ones at two different locations.
Hence understanding cities at all means understanding increasing returns to scale. But, says Krugman, increasing returns to scale is precisely what neoclassical economics doesn't know how to handle. My intuition here is even hazier. Krugman refers a few times to "unexploited economies of scale" causing problems for neoclassical economists, which suggests to me that there's some kind of arbitrage principle at work: in a perfectly competitive economy, the theory probably says somehow that factories would eventually scale up to the point that they're working in a constant-returns regime. Again, my intuition on this is hazy, but that's what the context of Krugman's writing suggests.
So if you're going to model city development, you need to model increasing returns. And if you're going to model increasing returns, you can't be talking about a perfectly competitive market. Apparently you're forced into a monopolistic competition model. These are just the sort of models that economics has always had a hard time understanding, says Krugman.
Krugman spends a good fraction of the book explaining this economic issue, so in a lot of ways "Development" turns into meta-economics: a study of why economics as a discipline behaves the way it does. Economics, says Krugman, ignored city development for so long because it didn't know how to model it; a certain standard of rigor has prevailed in economics since the 1960's, such that anyone who had nice ideas in prose but couldn't express them in mathematics was the proud owner of a dead letter.
So "Development, Geography and Economic Theory" is three things: a collection of toy models, a unification and deepening of some earlier work in economic geography, and a meditation on the value of those very models.