The key to the success of AI agents within a company? Shared memory and context.
This, according to Asanas CPO Arnab Bose provides detailed history and direct access from the start – with guardrail checkpoints and human oversight, of course.
This way, “when you assign a task, you don’t have to re-provide all the context of how your business works,” Bose said at a recent VB event in San Francisco.
AI as an active teammate, rather than a passive add-on
Asana launched Asana AI Teammates last year with the philosophy that, just like humans, AI agents should be connected directly to a team or project to create a collaborative system. To pursue this mission, the project management company has fully integrated with Claude from Anthropic.
Users can choose from 12 predefined agents (for common use cases like diverting IT tickets) or create their own, then assign them to project teams and immediately provide a history of which tasks have already been completed and which remain to be resolved. Agents also have access to third-party resources like Microsoft 365 or Google Drive.
“When this agent is created, it doesn’t act on behalf of someone, it shows up as a teammate and it gets the same sharing permissions, it inherits them,” Bose explained. Everything anyone does – including humans and AI – is documented to allow for “ease of explainability” and a “highly transparent and trustworthy system”.
But just like human workers, AI agents are controlled: workflows incorporate checkpoints, where humans can give feedback and ask the agent to tweak elements of a project or adjust research plans. This is documented in what Bose called “a very human-readable manner.”
It’s also important to note that the UI provides instructions and knowledge about agent behavior, and approved administrators can pause, modify, and redirect models in the API when they take actions based on conflicting instructions or start acting “in a strange way.”
“The person with edit rights can remove items that are in conflict and revert to the correct behavior,” Bose said. “We rely on this common human-understandable interaction model.”
Overcoming authorization, onboarding challenges
But because AI agents are very new, many challenges remain when it comes to security, accessibility, and compatibility.
Asana users, for example, must go through an OAuth flow and grant Claude access to Asana through their MCP and other public APIs. But getting all knowledge workers to know that this integration exists – and, more importantly, which OAuth subsidies are acceptable and which should be avoided – can be a daunting challenge.
Some of the challenges with direct cross-app OAuth grants could be centralized by identity providers, Bose noted, or by a centralized list of approved enterprise AI agents with their skills, “almost like an active directory or universal agent directory.”
However, currently, beyond what Asana does, there is no standard protocol around sharing knowledge and memory, Bose said. His team has received “many interesting inbound requests” from partners who want their agents to operate on the Asana work graph and benefit from shared work.
“But because the protocol or standard doesn’t exist, today it has to be a very personalized, tailored conversation,” Bose said.
Ultimately, the CPO currently calls three “extremely interesting” questions in AI orchestration:
-
How to create, manage and secure an authoritative list of known approved AI agents?
-
How can you enable application-to-application integrations as an IT team without potentially configuring dangerous or harmful agents?
-
Today’s agent-to-agent interactions are very solo. Clouds can be independently connected to Asana, Figma, or Slack. How can we finally achieve a unified, multi-stakeholder result?
The growing adoption of Modern Context Protocol (MCP) – the open standard introduced by Anthropic that connects AI agents to external systems in a single action, rather than custom integrations for each pairing – is promising, he noted, and its widespread adoption could pave the way for exciting new use cases.
However, “I think there’s probably no silver bullet right now,” Bose said.
