Agent-based systems suitable for dealing with applications where the environment is both dynamic and populated with competitors demand for sophisticated characteristics including adaptation, negotiation and co-ordination. We here briefly summarize some proposals on agents' negotiation capabilities including adaptation through reinforcement learning as well as qualitative multi-criteria negotiation and coalition formation protocols. Also, and inspired by robosoccer domain, some basic hints on knowledge representation for agents' team work are here described. All those proposals on automatic negotiation have been implemented through agent-based systems for different application domains (MACIV, SMACE, ForEV).