In this paper we present a multi-agent architecture for trading in electronic markets with asynchronous and related auctions. This architecture enables the development of a multi-agent system for a highly competitive environment, where all participants are competing for a limited number of goods. We define intelligent agent roles that tackle sub problems of trading, and present a solution for combining these results in a distributed environment. The agents’ typical tasks are price prediction, bid planning, good allocations, negotiation, among others. We use the Trading Agent Competition (TAC) environment as a case study to illustrate the suitability of our approach. We also present LearnAgents, a multi-agent system based on our architecture that achieved the third place in the 2004 TAC Classic competition.