Within three weeks of September, Meta, Microsoft and OpenAI each introduced persistent AI agents that can work on cloud computers long after users walk away from their own machines. The announcements, analyzed by Stephanie Nivinskus, principal at Ragan's Center for AI Strategy, give communications teams a reason to reassess what they delegate — even as enterprise access remains uneven.

Nivinskus argues that most comms professionals still equate AI with a chat window. Since ChatGPT launched, the dominant workflow has been to write a prompt and refine the output. "That is accurate. It is also the equivalent of describing a smartphone as something you use to make calls," she wrote.

What makes persistent agents different?

A chat answers questions. An agent takes an assignment and uses tools to carry out the steps. The differentiator for persistent agents is context retention: they hold the assignment's context and resume work without another prompt. Depending on the product and permissions, these agents can check for updates, follow up, or pick up an assignment days later.

The practical appeal for comms is continuity. Nivinskus offers a scenario: hand a persistent agent an assignment hours before leaving for vacation, return a week later, and find a months-long project kept moving while you were away.

A concrete example: a product announcement awaiting input from legal, an executive and a subject matter expert. With approved access and instructions, a persistent agent could track incoming feedback, flag conflicting edits and follow up with authorized reviewers. If the launch date changes, it could propose a revised review schedule. Humans still decide what gets published, but no one has to manually chase every update.

Who actually gets access?

Availability varies sharply by vendor:

• OpenAI: persistent agents called "dots" are in beta for eligible enterprise customers, but administrators must enable access.
• Microsoft: Autopilot is available only to selected Copilot customers for testing.
• Meta: Muse launched as a consumer app; the release does not establish enterprise availability.

How should comms teams prepare?

Nivinskus recommends working with IT on three steps before delegating anything.

1. Fix the workflow first. Identify where a recurring assignment stalls. Clarify who reviews and approves it before exploring automation. Work with IT to assess whether an approved tool supports the assignment and its required information.
2. Agree on authority and accountability. Comms defines which editorial decisions require human approval; IT establishes technical access and controls. Tracking approvals does not automatically authorize accepting edits, changing deadlines or publishing. Assign a comms owner to supervise the work.
3. Evaluate a small pilot together. Test an approved assignment for accuracy and follow-through. Measure the effort required to supervise it, including correcting mistakes. Use those results to decide whether to expand its role.

The September launches signal a shift in how the major AI vendors position their products: from conversational tools to delegated workers. For comms leaders, the near-term task is not adoption but preparation — mapping which recurring assignments justify delegation, and building the approval structures that make delegation safe before access widens.