Adam Smith thought the key to raising productivity was specialization and the division of labor, but this explanation has been marginalized in economic theory because it doesn't fit into neoclassical models. Using agent-based simulation (in particular an adapted version of the Howitt-Clower model of money) to step outside the constraints of traditional Walrasian equilibrium, I show how allowing for an endogenous division of labor vindicates several explanation of long-run growth which neoclassical analysis seems to rule out. For example, it is not the case, as the Solow model purports to demonstrate, that growth based on capital accumulation has a well-defined upper bound due to diminishing returns and depreciation, except in the special case where population and capital previously sufficed to exhaust the scope for productivity-enhancing specialization. In other cases, capital accumulation triggers the introduction of new goods, which were technologically known but not economically viable, and there is no particular limit to how much growth might be achieved in this way. The model provides a template for multiple candidate explanations for the Industrial Revolution: higher savings rates, larger markets due to international trade, or one or two new inventions, could all suffice to ignite a virtuous cycle in which exploration of the goods space expands the division of labor, which raises productivity and drives further exploration of the goods space, both from the demand side (larger markets make it easier to cover fixed costs of producing a new good) and the supply side (more capital is available to overcome the fixed cost hurdle). The model follows Romer (1990) in interpreting technological change as introduction of new goods; complements Romer (1990) by showing why the state of technology might differ across countries even when they have access to the same knowledge; and rivals Romer (1990) by opening up the possibility that the costs of invention might not be an important factor limiting growth via the introduction of new goods. (The alternative story is that it's easy to invent goods once they've become economically viable.) All results are derived from a simulation software developed by, and available from, the author.