Decision Processes in Agent-Based Automated Contracting

The Magnet system meets many of the challenges of modeling decision making for customer agents in automated contract negotiation. B usiness-to-business e-commerce is expanding rapidly, letting manufacturers both broaden their customer base and increase their pool of potential suppliers. However, negotiating supplier contracts for the multiple components that often make up a single product is a complicated process. Because component parts must be assembled and time dependencies often exist among operations , scheduling is a major challenge. Currently, there are no existing mechanisms or frameworks for automated negotiation and contracting among manufacturers , part suppliers, and specialized subcontractors. As the sidebar, " Related Work on Market-Based Architectures, " describes, current e-commerce systems typically rely instead on either fixed-price catalogs or auctions. However, such systems focus only on cost, which is just one factor in the complicated buyer-supplier relationship. For meaningful contract negotiations, systems must take into account other key factors, including schedule, quality, delivery performance, and flexibility. 1 The University of Minnesota has developed the Multi-Agent Negotiation Testbed (Magnet) system, 2 which is designed to support multiple agents in negotiating contracts for tasks with temporal and precedence constraints. Magnet has been under development since early 1998. Currently, its distributed market infrastructure is in place, as are the principal elements of its customer agents, which we focus on here. Magnet's customer agents have two key tasks. First, they must determine the specific contents of a Request for Quote (RFQ), which they send out at the start of negotiations to solicit supplier bids. The content of the RFQ determines how much time suppliers have to submit bids, and constrains the start and end times for the tasks. Next, once bids have been received, the customer agents review and