Macroeconomic models are typically solved through the imposition of a top-down general equilibrium solution constraining agents' rational be- havior. This is customarily obtained by recurring, explicitly or not, to the Walrasian auctioneer (WA) artifice. In this paper we aim at contributing to the small but burgeoning literature that deals with the consequences of removing it from the start by means of agent-based techniques. We let the textbook full-employment neoclassical macroeconomic model be populated by a large number of bounded-rational, autonomous agents, who are repeatedly engaged in decentralized transactions in interrelated markets. We set up a computational laboratory to perform several exper- iments, whose designs di¤er as regards the way we treat learning on the one side, and the institutional arrangement determining who - between firms and workers - is bound to bear the risk associated to incomplete markets on the other one. We show that our fully decentralized multi- market system admits the possibility to attain the WA full-employment solution, but also that serious coordination failures emerge endogenously as learning mechanisms and institutional settings are varied.