AI Agents Are Becoming Part of the Workforce

September 29, 2026 · Keith Pasley

DECISIONFIRST ANALYSIS

Pillar: Enterprise AI Architecture & Security

Why this matters

Persistent agents change the risk model. An AI assistant typically acts while someone is interacting with it. An always-on agent can continue pursuing objectives across systems and applications. That makes identity, authorization, monitoring and accountability part of the enterprise AI architecture.

Decision it changes

Before deploying persistent agents broadly, organizations should define an agent operating model: ownership, identity, permissions, approval requirements, lifecycle, monitoring and escalation. Agent deployment should follow those decisions rather than determine them afterward.

Who should care

CIOs, CISOs, CHROs, enterprise architects, identity and access management teams, AI platform leaders, governance teams and business executives deploying agents into operational workflows.

What I would ask next

If an AI agent works for your company 24 hours a day, who owns it, what is it authorized to do, and who is accountable for the decisions it makes?

OpenAI just introduced Dots, always-on AI agents designed to continue working even when the person using them isn’t sitting in ChatGPT.

Each Dot operates with its own cloud computer and browser and can work across connected applications. OpenAI demonstrated agents handling ongoing, multistep work rather than simply responding to individual prompts. Reuters describes the launch as part of OpenAI’s push into enterprise AI and the growing market for autonomous agents. (Reuters)

I think this represents a much bigger shift than another AI assistant.

AI agents are becoming persistent participants in the workforce.

An employee has an identity. A role. Permissions. A manager. Defined responsibilities. Access that can be changed or revoked. Their actions can be audited.

Persistent AI agents are going to need many of the same controls.

The enterprise decision is no longer simply whether employees should have access to AI.

Organizations now need to decide:

Who can create an agent?

Whose identity does it operate under?

Which applications and data can it access?

How much authority can it exercise without approval?

Who owns the decisions and actions it takes?

What happens to the agent when the employee who created it changes roles or leaves the company?

And who is accountable when an always-on agent takes the wrong action while no human is watching?

OpenAI says Dots can work continuously toward assigned goals. That is useful precisely because the human doesn’t have to supervise every step.

But that changes the governance problem.

We’re moving from governing employees using AI to governing AI that performs work on behalf of employees.

That requires more than acceptable-use policies.

It requires agent identity, least-privilege access, authorization boundaries, approval thresholds, monitoring, audit trails, lifecycle management and a clearly identified human owner.

The companies that solve those questions will be in a much better position to scale agentic AI.

Because the real enterprise question isn’t simply:

What can the agent do?

It’s:

What should we authorize it to do, and who remains accountable for the outcome?

Technology doesn’t create business value. Better decisions do.

Source:

https://devday.openai.com/?trk=public_post-text&utm_source=chatgpt.com

Back to Insights