Abstract
This survey paper examines the recent advancements in AI agentimplementations, with a focus on their ability to achieve complex goals thatrequire enhanced reasoning, planning, and tool execution capabilities. Theprimary objectives of this work are to a) communicate the current capabilitiesand limitations of existing AI agent implementations, b) share insights gainedfrom our observations of these systems in action, and c) suggest importantconsiderations for future developments in AI agent design. We achieve this byproviding overviews of single-agent and multi-agent architectures, identifyingkey patterns and divergences in design choices, and evaluating their overallimpact on accomplishing a provided goal. Our contribution outlines key themeswhen selecting an agentic architecture, the impact of leadership on agentsystems, agent communication styles, and key phases for planning, execution,and reflection that enable robust AI agent systems.