The challenge of information overload in dealing with ever increasing variety and size of digital data on the Web is now receiving serious attention of the researchers. The information brokering architecture provides one approach to addressing issues at data, information and knowledge levels. While reasonable progress has been made in achieving system interoperability as well as syntax and structure level interoperability of data and information systems, the semantic level is the key to a more satisfactory solution. This paper discusses whether a multi-agent approach can help achieve semantic information brokering by supporting three of the capabilities needed: (a) extract and use semantic metadata descriptions from the underlying data; (b) handle information requests independent of the structure, format and media of the underlying data; and (c) share, exchange, and interoperate across collections of knowledge represented using multiple domain specific ontologies.