Microsoft ISE shares how it solved context handoff in multi-agent AI systems

Microsoft's Industry Solutions Engineering team has published a detailed account of how it managed conversational context across multiple independent AI agents using the Agent2Agent (A2A) protocol, an open standard now maintained by the Linux Foundation. The core challenge was passing coherent conversation history between coordinator and domain agents that share no common infrastructure or persistent memory. After evaluating three approaches — shared storage, per-agent stateful memory, and inline payload delivery — the team chose to embed summarised conversation history directly within each message. To control payload size, history is condensed every ten conversational turns, a tunable threshold the team acknowledges carries a risk of losing detail or introducing inaccuracies. The method also provides a security benefit, allowing the coordinator to selectively filter sensitive information before passing context to any given domain agent.