Asana's AI agents share company memory, not secrets

Enterprise teams developing AI agents face challenges with chatbots that lack memory of past interactions or tasks.

Asana has introduced a solution called Agentic Work Management (AWM). According to Chief Product Officer Arnab Bose, AWM treats AI agents as coachable teammates that operate alongside humans, differing from traditional one-to-one assistants.

AWM is built upon Asana's 18-year-old Work Graph architecture. This graph-based database organizes company information hierarchically, from tasks to company-wide goals. This allows the AI agent to leverage broader company context, track project statuses, and share information with human colleagues. Bose emphasized that the agent operates on shared, company-wide data, not just individual user files.

The system also incorporates robust security measures to protect confidential information. While agents share memory, the system prevents unauthorized users from accessing sensitive data, such as M&A projects. Additionally, AWM dynamically manages the selection of the appropriate AI model based on the task and offers users predictable pricing based on task completion rather than complex computation.

AWM is already in production with several customers, including FedEx. The solution aims to address the issue of statelessness in current enterprise AI solutions by creating a persistent state and recording metadata on task execution and its impact on company goals.