Combining object and meta level search using built in tasks

There are three major subtleties which may restrict a problem solver: How can it be designed so that it controls its own mechanism for making heuristic choices? If this control process is solved by the same mechanism which solves basic tasks, how can the problem solver avoid unlimited regression in a tower of control problems? When it solves one task by creating some new entity, how can this entity be made available for the solution of other tasks? This paper outlines answers. Every task, including every control task, may have associated tasks to control it; but these control tasks are created by a lazy process which only creates them when they are needed. The lazy creation process is itself encoded as tasks the built in tasks. All instances of tasks are stored in an agenda, and are chosen and interpreted with an augmented conventional architecture.