It is widely held that the vehicle for service convergence will be a QoS enabled Internet. Along with convergence and service integration comes the need for Quality of Service, QoS, assurances. Closely related to the provision of adequate QoS is the issue of pricing. Pricing for IP QoS is at present an open problem that must be addressed prior to wide-scale deployment of QoS in the Internet. The diversity of technologies, services of various qualities and multiple players has brought about the need for innovative models to better describe the interplay between resource allocation, quality of service and pricing schemes. The centralized planning approach of the past based on cost minimization for a specified demand and quality of service requirement, is gradually giving way to a decentralized, multi-player, multi-goal decision structure resembling commodity markets. We consider a multi-domain architecture, employing DiffServ/MPLS within administrative domains and Dynamic Service Level Agreements, DSLAs, linking these administrative domains to each other and to a centralized bandwidth broker. This seminar will review our approach and present recent results on two selected topics within this framework. The first topic, "Pricing for QoS provision across multiple ISP domains", a joint work with Soheil Saberi and Roland Malhamé, employs a game theoretic formulation for the price and throughput across multiple ISP domains. The second topic, "Adaptive routing for an NGI using decentralized reinforcement learning", is a joint work with Fariba Hiedari and Shie Mannor. We describe our QoS routing algorithm for a DiffServ enabled MPLS network and present and compare simulation results for reinforcement learning algorithms which compare favourably with AT&T's "Success To the Top", QoS routing algorithm.