First Look · AI Workflows · Updated June 25, 2026
Asana AI Teammates and Claude Managed Agents: What Changes for Project Managers
Two launches in June 2026 quietly rewired the project management stack. Asana shipped AI teammates, including 21 prebuilt agents that take assignments inside a project the same way a person does. Anthropic shipped Claude Managed Agents, which run multi-step work toward a defined outcome with less hand-holding. The headline for a senior PM is simple: the tool stopped being a place to track work and started being a worker you direct.
What launched and what it means, in plain terms: In June 2026 Asana introduced AI teammates that sit inside a project, hold a role, and can be assigned tasks, with 21 prebuilt teammates available out of the box and a beta for building custom ones. In the same window Anthropic released Claude Managed Agents, which coordinate several steps of work toward a measurable outcome rather than answering one prompt at a time. For project managers this shifts the job from pushing tasks through software to briefing, supervising, and quality-checking a small set of agents that draft status updates, chase blockers, and summarize decisions. The coordination toil shrinks. The judgment work grows.
Key Takeaways
- Asana AI teammates can hold a role and take assignments inside a project; 21 come prebuilt, with custom teammates in beta.
- Claude Managed Agents run multi-step work toward a stated outcome, so the unit of delegation is now a result, not a single message.
- The PM job moves up the stack: from chasing and compiling status to writing clear briefs, setting guardrails, and reviewing agent output.
- Agents handle coordination; they do not own accountability, stakeholder trust, prioritization tradeoffs, or the call on ambiguous risk. Those stay human.
The shift: the coordination toil is being automated
Most of a project manager's week never appears on the roadmap. It is the connective tissue: pinging six people for an update, stitching their replies into a status note, noticing that the design review slipped and quietly rebooking it, turning a messy meeting into action items, and writing the same decision summary three different ways for three audiences. None of that is strategy. All of it is necessary, and all of it eats hours.
That connective tissue is exactly what the new agent model targets. An Asana AI teammate can be assigned to "keep this launch tracker current," and it pulls status, flags slippage, and updates fields without a human relaying every message. A Claude Managed Agent can be handed an outcome like "produce a stakeholder-ready weekly summary from these threads and the project board," and it works through the steps to get there instead of waiting for the next instruction.
The PM's value was never the status update. It was the judgment behind which work mattered and which risk to escalate. The agents just took the part that was never the point.
This is the real story under the launch noise. Project software spent fifteen years getting better at storing tasks. The 2026 generation is getting better at doing them. When the place you tracked work becomes a teammate who does work, the human in the loop has to move to where the leverage actually is.
What a PM's job becomes
The role does not disappear. It changes shape. The senior project manager who thrives in this stack spends less time as a router of information and more time as a director of agents and a steward of judgment. Four responsibilities move to the center.
1. Briefing instead of relaying
An agent is only as good as the brief it receives. The skill that used to be optional, writing a crisp, unambiguous instruction with success criteria and constraints, is now the core skill. A vague brief produces confident, wrong output at scale. A sharp brief produces a draft you can ship after a quick edit.
2. Supervising and reviewing
When an agent drafts the status note or chases the blocker, the PM becomes the editor and the final check. The job is to catch the agent's misread, the stale assumption, the tone that will land wrong with a nervous executive. Review is faster than authoring, but it is not free, and it is where errors get caught before they reach a stakeholder.
3. Prioritizing and sequencing
Agents are good at executing a defined task. They are weak at deciding which of five competing tasks deserves the team's next week. Tradeoffs between scope, deadline, and risk are still a human call, informed by context the agent does not have: the politics, the unspoken executive priority, the customer who will churn if this slips.
4. Owning accountability
When a project lands or fails, a person answers for it. You cannot delegate accountability to an agent, and no stakeholder will accept "the teammate got it wrong" as an explanation. The PM remains the single throat to choke, which is precisely why the supervision and review work cannot be skipped.
A what-to-do-now playbook
You do not need to re-platform your team this quarter. You need to test the new model on real work and build the habits before the tools become table stakes. Here is a sequence that works.
- Pick one low-risk, high-toil job to delegate first. Weekly status compilation is the ideal candidate: repetitive, time-consuming, and easy to verify. Assign it to an Asana AI teammate or hand the outcome to a managed agent, and compare its draft to what you would have written.
- Write the brief like a contract. State the outcome, the audience, the format, the source material, and the constraints. "Summarize the launch project for the exec team in five bullets, flag any task more than three days late, do not speculate on causes." Specificity is the whole game.
- Keep a human review gate on anything a stakeholder sees. Let the agent draft. You approve. Never let unreviewed agent output go straight to a client, an executive, or a performance record.
- Log what the agent got wrong. For two weeks, note every miss: the wrong priority, the stale status, the misjudged tone. That log becomes your guardrail list and tells you which jobs are not ready to delegate yet.
- Reinvest the reclaimed hours in the work agents cannot do. The point of automating compilation is not to do nothing. It is to spend the recovered time on stakeholder alignment, risk anticipation, and the prioritization calls that decide whether the project actually succeeds.
Guardrails before you delegate
Set these guardrails first
- Confidentiality and access. An agent assigned inside a project can read what the project contains. Confirm what data it can see and where its outputs go before you point it at anything sensitive, especially HR notes, legal matters, or unreleased financials.
- No unsupervised stakeholder output. Agent drafts are inputs to your judgment, not finished communications. A confident wrong summary sent to a board is worse than a slow human one.
- Watch for automation drift. An agent that quietly rebooks meetings or changes statuses can mask a real problem instead of surfacing it. Keep visibility into what the agent changed and why.
- Accountability stays with the human. Document that you reviewed and approved agent-assisted decisions that carry consequences. The audit trail is yours, not the tool's.
What AI does not replace
The agents took the coordination, not the leadership. A project manager still owns the things that have no clean prompt: reading the room when an executive is unhappy and will not say it, deciding which deadline to defend and which to sacrifice, holding a team's trust through a hard quarter, and absorbing accountability when something breaks. Agents make a good PM faster. They do not make a weak one good, and they do not make the human optional. The professionals who win this transition treat the agent as the most capable junior teammate they have ever managed, and they manage it.
How this maps to a comparison you can use
| Project work | Who does it now | Who owns the outcome |
|---|---|---|
| Compile weekly status | Agent drafts, PM reviews | PM |
| Chase blockers and updates | Agent, with PM visibility | PM |
| Turn a meeting into action items | Agent drafts, PM confirms | PM |
| Decide what to build next | Human | PM |
| Escalate ambiguous risk | Human | PM |
| Answer for the result | Human | PM |
Frequently asked questions
Will Asana AI teammates and Claude Managed Agents replace project managers?
No. They automate the coordination work, compiling status, chasing updates, drafting summaries, that consumed much of a PM's week. They do not own prioritization tradeoffs, stakeholder trust, or accountability for outcomes. The role shifts from routing information to directing agents and exercising judgment.
What are the 21 Asana AI teammates?
Asana's June 2026 launch includes 21 prebuilt AI teammates that can hold a role inside a project and take assignments, plus a beta for building custom teammates. They handle defined, repeatable project tasks so the human can focus on review and direction.
What is a Claude Managed Agent?
It is an agent that runs multi-step work toward a stated outcome rather than answering a single prompt. Instead of asking it one question at a time, you give it a result to achieve and constraints to respect, and it works through the steps to get there. For PMs, that means delegating an outcome like a stakeholder-ready summary, not just a single query.
What should a PM delegate to an agent first?
Start with weekly status compilation: it is high-toil, repetitive, and easy to verify against what you would have written. Keep a human review gate on anything a stakeholder will see, and log what the agent gets wrong to build your guardrail list.
What should a PM never automate?
Never automate the final call on competing priorities, the escalation of ambiguous risk, sensitive communications to executives or clients, anything touching confidential HR or legal data without review, and accountability for the result. Those remain human work.
Part of TLY's AI Workflows
This is a First Look in TLY's AI Workflows, where we test how senior professionals actually use AI and report honestly on what works.
Go deeper: The AI for Managers course teaches the weekly operating system for directing agents and keeping judgment human. Join The Leverage Club ($49) for the practitioner community and ongoing playbooks. Not sure where you stand? Take the 2-minute AI leverage quiz.