Asana has introduced shared memory capabilities for AI agents that operate across enterprise organisations, addressing a critical gap in how agentic systems retain and leverage institutional knowledge. The feature allows multiple AI agents deployed across different teams and departments to access a unified memory layer, enabling continuity in decision-making and reducing the redundancy of agents operating in isolation. This moves beyond the stateless nature of most current AI implementations, where each interaction exists independently and agents cannot build on previous conversations or outcomes. For CX teams already managing multiple support channels through platforms like Zendesk or Freshdesk, this represents a fundamental shift in how agent-assisted workflows could function—agents handling billing inquiries could theoretically reference patterns from previous customer interactions across the entire support organisation, rather than treating each ticket as a discrete problem.
The critical tension here lies in the security architecture underpinning this shared memory. Asana's approach appears to compartmentalise access based on organisational permissions, meaning agents can share institutional memory without exposing confidential customer data or sensitive business logic across unauthorised teams. This distinction matters significantly for CX operations, where support agents need visibility into customer history without accessing financial records or proprietary product roadmaps. The question becomes whether this permission-based memory model can scale reliably as organisations grow—can a support team lead confidently deploy agents that access shared memory without inadvertently creating data leakage vectors, particularly when agents themselves are increasingly autonomous in their decision-making?
The broader implication is that agentic AI in CX is moving from isolated point solutions toward integrated knowledge systems. Teams currently evaluating agent platforms should assess whether their chosen vendor has a credible answer to the memory and continuity problem, as this capability will likely become table stakes for enterprise deployments. Organisations with fragmented support stacks across multiple vendors face a harder integration challenge, whilst those consolidating around platforms with native agentic capabilities gain a structural advantage in building genuinely intelligent support operations.
Enterprise teams building AI agents keep hitting the same wall: a chatbot that can answer a prompt but can't remember what the last five people asked it, and can't tell you whether last month's version actually worked.In a fireside chat with VentureBeat's Sam Witteveen at VB Tran