CHAPTERS
- 0:14 – 0:45
Asana’s “Agentic Enterprise” vision: AI teammates as real actors in work
Arnav frames Asana’s goal as moving from individual AI usage to an enterprise model where AI agents collaborate with people on complex, multi-step outcomes. The key idea is that agents should behave like teammates inside the system, not external chat tools.
- •Agents as first-class actors embedded in enterprise workflows
- •Human + agent collaboration to complete approvals and end-to-end processes
- •Focus on real-world outcomes across functions (IT, ops, product, marketing)
- •AI teammates in Asana are generally available (as of March)
- 0:45 – 1:46
Why “single-player agents” fall short: no compounding knowledge or shared memory
He contrasts common enterprise experimentation with agents against Asana’s target state. The problem: agents are often used in isolated, one-off interactions that don’t build organizational knowledge or support true multi-user workflows.
- •Most companies use agents in a handoff-based, single-user pattern
- •Lack of compounding knowledge and shared enterprise memory
- •Missing “multiplayer” human-in-the-loop collaboration
- •Motivation for Asana’s approach to shared, durable agent context
- 1:46 – 2:47
Multiplayer mode + enterprise memory: agents that keep improving with team use
Arnav explains how shared usage and durable memory make agents more valuable over time. He illustrates this with an internal competitive intelligence agent that remains useful even after its creator left the company.
- •Agents can be shared across users with enterprise controls
- •Enterprise memory accumulates from historical interactions
- •Example: competitive intelligence researcher agent used post-employee departure
- •Shared usage drives continuous improvement and reuse of prior context
- 2:47 – 3:48
Context and governance: decisions, approvals, and auditability as agent inputs
Asana emphasizes that agents need deep organizational context—who does what, why decisions were made, and how approvals happened. This is delivered with security guardrails, RBAC, and auditability to enable trusted action-taking.
- •Agents ingest context like historical decisions, campaign briefs, approvals
- •Security, guardrails, and audit trails are core requirements
- •Goal: enable real action, not just suggestions
- •Asana positions itself as a “system of action” for agent-driven work
- 3:48 – 4:49
The Asana Work Graph as the binding layer for human + agent collaboration
He describes Asana’s long-built work graph—goals, portfolios, projects, tasks, workflows—as the structured context agents can use. This enables both a human-friendly UI and an agent-friendly representation of how work gets done.
- •Work graph hierarchy: goals → portfolios → projects → tasks/workflows
- •Designed over many years as enterprise context infrastructure
- •Agents can operate across multiple people while respecting access boundaries
- •Foundation for “true multiplayer” agent behavior
- 4:49 – 6:49
Why Claude Managed Agents: multi-step execution, verification loops, and parallelism
Arnav introduces where Claude Managed Agents fits: running the multi-step workflows reliably and with quality controls. He highlights prototyping speed, built-in verification, and enabling multiple agents to work in parallel.
- •Managed Agents used to complete complex multi-step actions
- •Faster prototyping vs building manual agent loops
- •Built-in verification loop and grader improve output quality
- •Parallel agent workstreams support knowledge-worker planning tasks
- 6:49 – 7:19
From Messages API to Managed Agents: what improved for Asana’s build
He compares Asana’s prior approach using the messages API with the newer Managed Agents setup. Key gains include not having to build infrastructure for loops/files/execution and improved verification quality.
- •Reduced need to build manual loops, file management, code execution
- •Better built-in verification and quality assurance
- •Enables multiple agents operating independently in parallel
- •Lets Asana focus on UI/coordination layer and enterprise security controls
- 7:19 – 8:21
What AI teammates can do today: 21+ prebuilt roles and connected integrations
Arnav outlines the current catalog of AI teammates tailored to different enterprise personas. These agents operate within Asana constructs and can also interact with connected tools to produce artifacts and take actions.
- •21+ prebuilt AI teammates mapped to key enterprise functions
- •Use cases: launch planning, specs, goals, resourcing/capacity planning
- •Operate within portfolios, timelines, and other Asana objects
- •Integrations: Google Drive, Microsoft 365 (and more upcoming)
- •Agents can generate artifacts like slides, comments, and HTML
- 8:21 – 9:52
Internal dogfooding example: a “product org thought buddy” for cross-team feedback
He shares how Asana uses AI teammates internally to scale expert feedback. Marketing can assign a task to a product-team agent to generate context-aware, accurate feedback that the whole product org can review and refine.
- •Agent holds product strategy, roadmap context, and past trade-offs
- •Marketing assigns agent a task for keynote speech feedback
- •Outputs are visible to the whole team for reaction and nudges
- •Memory persists across runs, improving future responses
- 9:52 – 10:52
Demo walkthrough setup: campaign brief + landing-page prototype as a multi-step job
Arnav transitions into a demo video showing a marketer launching a campaign. The workflow requires generating a campaign brief and producing iterative landing-page mockups, illustrating multi-step agent execution.
- •Scenario: marketer needs a brief plus landing page prototype
- •Starts from a task on an Asana Kanban board
- •Uses prebuilt AI teammate from the teammate gallery
- •Demonstrates multi-step generation (document + HTML)
- 10:52 – 12:53
Demo details: auto-context from the work graph and Managed Agents grading in console
The demo shows the teammate pulling relevant prior projects/portfolios into memory to improve outputs. It also shows the Claude console runs where the outcome and grading/verification loop are executed.
- •Teammate automatically selects relevant work graph objects for memory
- •Generates both campaign brief content and an HTML landing-page mockup
- •Claude console shows Managed Agents runs and grading iterations
- •Asana combines RBAC/context with Managed Agents quality assurance
- 12:53 – 15:56
Multiplayer iteration, audit trail, and RBAC ownership controls for agent memory
Arnav highlights how multiple teammates can comment to iterate the work (e.g., color/theme changes, more minimal design). All interactions are auditable, can be routed to approvals, and agent sponsors can manage or delete memories.
- •Feedback via comments drives iterations (e.g., switch primary color to blue)
- •Agent memory updates so future users don’t repeat mistakes
- •Second reviewer requests a more minimalistic version (multiplayer)
- •All prompts/responses are captured for auditability and approvals
- •RBAC defines agent owners/sponsors who can manage access and memory
- 15:56 – 17:59
What’s next: proactive agents, broader workflows, and enterprise-scale planning
He closes the product direction: larger workflows like full launches, capacity planning, dashboards, risk reports, and proactive issue detection. The goal is agents that can advance work even without being explicitly assigned a task.
- •Expand to full launch planning and enterprise-wide resource management
- •Generate dynamic dashboards and risk reports
- •Proactive alerting and recommended remediation actions
- •Learning from team patterns to improve skills and outcomes
- 17:59 – 24:56
Q&A: verification contract, rubrics, skill maintenance, and third-party integrations
Audience questions probe how Asana injects domain context into verification, how rubrics are designed, how skills are maintained over time, and how third-party tools are integrated. Answers emphasize outcome definitions with context, prompt-like rubric iteration, shrink-wrapped skill strategy, and integrating at both Asana and MCP/Managed Agents levels.
- •Verification: pass Asana context in the outcome definition; internal QA + human-in-loop improvements
- •Rubric design: treat grader instructions like prompts; instrument and iterate evaluations
- •Skill maintenance: Asana ships prescriptive, shrink-wrapped skills; possible future customization
- •Integrations: handled both in Asana’s agent loop and at the MCP level with Managed Agents
