This talk covers in depth only one-though mostly important - part of the Internet middleware (IMW) - policy, and provides some in-depth study of one aspect of it - a policy-based programming of network elements to support advanced Internet services. We start however with other IMW parts as a background constraining and motivating the agenda. The talk will be concluded by a discussion of scalability design methodology for the policy-based programming. First, following popular but hard to accept practice, we introduce IMW as an unstructured set of services and resources between Internet application programming interfaces and the Internet Protocol. Then we examine IMW core components (APIs, AAA, Policy, Directories, resource management, discovery and retrieval, QoS, security, operational tools) and observe that they all are organically using a concept of a group and a concept of a policy. We conclude this introductory part with a conclusion that a combined use of middleware services and resources is a hot a research issue, providing several examples of research challenges. We then make a big picture of existing Internet as a patchwork of dedicated client-server protocols, while Internet services, especially those with call features (SIP, RSVP, RTSP, mobility, AAA) and QoS demands require coordinated behaviour (known as midcom - middle box communication) from several dedicated protocols, almost always including a policy protocol. To achieve required co-ordination one can either develop yet another client server protocol (that clearly does not scale to a number of emerging combined usage scenarios), or use application level middleware (explicitly non-goal for this talk), or extend the very notion of policy to meet the co-ordination requirement. This extension asks for a broader view on policy. Rather than plain “device configuration” [IETF, DMTF], a policy is “a rule that defines a choice in the behaviour of a system” [M. Sloman]. A policy-controlled component in turn needs to have Externalised Behaviour Choices (EBC). Once this is done consistently and safely, one can: separate EBC and policy rules design concerns, one can influence behaviours by changing rules, compose rules from components originating in multiple sources (combined use of IMW components), thus achieving policy-based self-organisation. We demonstrate a practical example of self-organising behaviour. Policy conflicts, similar to feature interaction, are detected mainly at policy enforcement points, while conflict resolution is possible mostly during policy computation or adaptation to an enforcement environment. We show how early conflict resolution can be done with yet another extension of a policy by a meta-data. Finally, we claim that policy programming together with event based interaction and group communication of IMW components enable design of evolvable systems. We show a snapshot of emerging design methodology that keeps complexity under strict control, yet allows very high level of flexibility. The tutorial will be concluded by an overview of an Internet research frontier.