We have implemented a learning interface agent for a commercial application in a real world domain, and have tested it with real world data. Results have shown that multi-agent collaboration steepens the agent's learning curve, and helps in new, unseen situations. Trust modeling allows each agent to build a model of each agent's area of expertise, and consult only those agents which will be useful for each area.