CHAPTERS
Asana’s “Agentic Enterprise” vision: AI agents as real teammates
Arnav frames Asana’s goal: move from one-off AI interactions to an “Agentic Enterprise” where humans and AI agents collaborate on complex, multi-step work. Agents are positioned as first-class actors in the work system, able to drive outcomes like approvals and end-to-end workflows.
Why most enterprise agent usage is still “single-player”
He contrasts today’s typical agent usage—an individual gets an output and hands it off—against true collaborative workflows. This single-player pattern lacks compounding knowledge, shared memory, and multi-human governance.
Enterprise memory and shared context: agents that improve over time
Arnav describes how agents can retain institutional knowledge through ongoing use, becoming more valuable as more people interact with them. He shares an anecdote about a competitive intelligence agent built by a former employee that continues to be used and improved.
Context + governance: RBAC, auditability, and “system of action” foundations
He explains that Asana provides rich work context (decisions, approvals, project history) while maintaining security and guardrails. Asana’s work graph and controls enable agents to act safely like onboarded teammates.
Where Claude Managed Agents fits: powering multi-step skills and outcomes
Arnav outlines how Claude Managed Agents enables complex, multi-step workflows such as generating a campaign brief plus landing page mockups (HTML). Managed Agents helps Asana focus on its differentiators while delegating iterative quality control to Anthropic tooling.
Why Managed Agents beats a DIY loop (vs Messages API)
He compares the prior approach (Messages API) with Managed Agents, highlighting the engineering burden of manual agent loops and tooling. Managed Agents simplifies file management/code execution patterns and supports parallel agent work for knowledge tasks.
Current product scope: 21+ pre-built AI teammates and integrations
Arnav describes Asana’s set of pre-built AI teammates designed around ideal customer profiles across functions. These teammates can take actions inside Asana and pull/push work across external tools (e.g., Google Drive, Microsoft 365).
Internal dogfooding example: “product org thought buddy” feedback loop
He shares how Asana uses an internal agent that contains product strategy and tradeoff context. Marketing can assign it a task for product input (e.g., keynote draft feedback), and the output becomes a shared artifact the team can refine over time.
Demo walkthrough: marketer creates campaign brief + landing page prototype
The demo shows a marketer starting from a Kanban task, selecting a pre-built teammate from the gallery, and automatically pulling relevant work graph context. The agent produces both a campaign brief document and an HTML landing page mockup.
Inside the Claude console: outcomes, runs, and the verification/grader loop
Arnav briefly shifts to what this looks like in the Claude console. Asana passes an “outcome” definition, and Managed Agents performs iterative grading/verification to reach an acceptable final output.
Iteration + memory: changing requirements get retained for future users
He demonstrates how simple comments drive iterative improvements (e.g., changing a primary color to blue). Crucially, these nudges can be incorporated into the agent’s memory so future users benefit and the same mistakes aren’t repeated.
Multiplayer collaboration, audit trail, and agent ownership controls
A second teammate joins to request a more minimal design iteration, showcasing multi-user collaboration with the agent. Arnav emphasizes that the entire interaction history is logged for auditability, and agent managers can govern memories and access scope.
What’s next: deeper workflows and proactive agents
Arnav outlines future directions: larger multi-step workflows (launch planning, capacity planning), automated dashboards and risk reporting, and proactive agents that can “wake up” to help without explicit assignment. The goal is to learn from team patterns and move work forward earlier.
Audience Q&A: verification contract, rubrics, skill maintenance, and integrations
The Q&A covers how Asana injects context into outcome definitions and layers additional internal QA, plus how rubric/grader design resembles prompt iteration. They also discuss a shrink-wrapped approach to skill maintenance (for now) and integrating third-party tools both within Asana’s loop and via MCP with Managed Agents.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome