Agents you can coach: how Asana builds human-agent teams with Claude

Hi everyone, it's Shii-chan! Today's story is about how Asana turns AI agents into members of the team.

Claude Blog claude.com

What was announced?

Claude Blog published a post in which Asana's CPO, Arnab Bose, explains how Asana builds human-agent teams on its Work Graph platform, with defined roles, shared memory, and visible workflows. It is a case study and a set of ideas rather than a product launch.

Why it matters

When a company uses agents, questions come up: what may they do, who teaches them, and can anyone see what they did? The post answers with four principles Asana relies on.

Dive Deep

The four principles:

• Structure the work first: employees consolidate unstructured data (Slack, Zoom, docs) by discussing it with Claude, log actionable items into Asana's projects and tasks, and agents then act within that structure
• Role-based access and governance: each agent has a defined purpose, scoped permissions, and named administrators. Access is bounded by the permissions of the person who triggers the agent. Agents are assigned roles such as content writer, project manager, or analyst, with pre-built skills
• Separate training from usage: anyone can give feedback, but only admins and editors can change permanent memory. For example, the communications team trains the writing agents
• Make work visible in shared tasks: research plans and steps are posted in shared tasks, and reviewers can comment and steer. The request, iteration, and output stay in one place

Three examples are shown:

• Slack channel support agent: answers product questions, flags documentation gaps, and routes to product intake
• At-risk renewals digest: a daily briefing for executives on the customer portfolio
• Engineering cycle planning: agents populate planning boards from customer feedback, and managers query outcomes through Command by Asana

Wrap-up

• Give agents roles, scoped permissions, and administrators
• Permissions are capped by the person who triggers the agent
• Keep the right to teach separate from the right to use
• Show the work in shared tasks so people can coach the agents

This is for anyone designing how to roll out agents in an organization!