Building intelligent software agents using traditional knowledge acquisition methods is difficult because of the dynamic properties of Internet. Knowledge about Internet can be divided in two levels. The first level is knowledge about tools and methods for searching information in Internet. The second level is knowledge about information in Internet (meta level). We focus on the meta level. Our approach is to identify knowledge sources in Internet. The knowledge may be implemented in a software agent and the planning can be improved by combining the knowledge from several intelligent software agents.