- Published on
Asana's AI agents share memory across your company β but not your secrets
- Authors
- Name
- DeepHorus
Asana unveiled its new Agentic Work Management (AWM) system at VB Transform 2026, with Chief Product Officer Arnab Bose detailing how AI agents can share enterprise-wide memory without compromising confidential information. The system treats AI agents as coachable teammates working alongside humans, rather than one-to-one assistants. This marks a significant shift for companies struggling with stateless chatbots that cannot remember past interactions or understand broader company context.
AWM is built on Asana's 18-year-old Work Graph architecture, a graph-based database organizing tasks into projects, portfolios, and company-wide goals. The structure, called the Pyramid of Clarity, enables the AI to trace how a delayed design task impacts a corporate revenue goal. Bose explained that the agent works off a shared ledger for the whole company, not just a single prompt or local file system.
The product is already in production, with several customers live on it, including FedEx, which published its own case study. However, shipping AWM required solving critical technical hurdles. The primary challenge is data governance: if an executive builds workflows for a confidential project, the system must prevent memory from leaking to unauthorized employees later interacting with the same agent.
Asana engineered access controls to govern what triggers memory creation versus simple task execution. The system also implements dynamic model routing, automatically sending complex tasks to frontier models like Anthropic's Opus or OpenAI's models, while lighter tasks get down-leveled to faster, cheaper options. This abstraction frees knowledge workers from crafting perfect prompts, making the experience feel like assigning work to a human.
The third challenge is billing abstraction for agentic tasks, ensuring costs are managed as agents operate autonomously. Bose emphasized that the agent's shared memory and real-time view of company objectives enable it to update project statuses and coordinate cross-functionally. This approach provides a blueprint for developers building agentic systems at scale, addressing bottlenecks in enterprise context and security.
Looking ahead, Asana's AWM demonstrates a viable path toward true human-agent teamwork in the enterprise. By leveraging a mature data architecture and robust guardrails, the company is positioning itself as a leader in the emerging agentic operating system space. The implications are significant for productivity and data security as more organizations adopt AI teammates.
