Six weeks after Muse, every AI agent got a job. Almost none got a manager.

On September 8, 2026, Meta launched Muse, a personal AI agent that sends email, books travel, fills out forms, lowers bills and buys things through Link by Stripe. You connect your services one at a time and it goes to work.

TechCrunch put the bet plainly. An agent like this is "one that requires significantly more trust than social media ever did."

Muse was the consumer opening act. In the six weeks since, nearly every large AI company shipped an agent built to work for a business, run for hours or days, and act without someone watching each step.

What shipped

Date	Company	What it is
Sep 8	Meta	Muse, a personal agent that acts across your apps and pays through Link by Stripe
Sep 28	Salesforce	Agentforce Coworker, described as "your autonomous AI teammate," plus Agent Health Monitoring for production agents
Sep 29	OpenAI	Dots at DevDay: always-on agents that "take on responsibilities on their behalf" and connect to more than 4,000 apps
Sep 30	Cloudflare	Monetization Gateway beta, so websites can charge agents per request
Oct 8	Google Cloud	Gemini agent, "a single, universal agent for work" whose workflows can run for hours or days
Oct 15	Meta	Muse Spark Open, the model under Muse, released with open weights

Look at the verbs. The agent is no longer answering a question someone typed. It is taking on a responsibility, running a workflow for days, working for a team or a role. Google says its agent can act as a personal assistant or as a member of a team doing a job such as finance analyst.

That is a job description. And a job has an owner, a measure and a review. The releases above describe what the agents can do. None of them says who answers for what the agent did on Tuesday.

The gap, in Gartner's numbers

Gartner released two surveys today at its IT Symposium. Read them together.

71% of organizations expect to deploy agentic AI within the next 12 months.
21% have agentic AI governance capabilities in place today.
4% of CIOs have plans to redesign roles for a human-agent workforce.
13% of CIOs say their enterprise has realized significant or transformational value from AI.

The first two numbers come from a Q2 2026 survey of more than 2,000 CIOs and technology executives. The last two come from Gartner's survey of 2,002 CIOs and technology executives, fielded May to July 2026.

Most companies will have agents next year. Only about one in five will have the rules to govern them. And only 4% plan to change the jobs around them.

Why governance is not the word for it

"Governance" sounds like a policy binder and a committee. That is not what goes missing when an agent misbehaves.

What goes missing is plain. Nobody can say whose agent it was. Nobody wrote down what it was supposed to produce this week. Nobody looked at the result in a meeting, and the thing it promised to do last month quietly stopped happening.

Those are accountability questions, and every company already has a way to answer them for people. There is a chart that says who sits in which seat. There is a number each seat is measured on. There is a weekly meeting where the numbers are read and the commitments are checked.

Agents arrived without a seat on that chart. They sit in an admin console, a vendor dashboard, a settings page. OpenAI's own enterprise default says something here: Dots is off for business workspaces until an admin turns it on. A switch tells you whether the agent runs. It does not tell you who owns its results.

The one question

Before your company turns on the next agent, answer one question for it.

Who owns what this agent does, and where does anyone check?

If the answer is a name and a weekly number, the agent can go to work. If the answer is "IT set it up" or "it's in the vendor's dashboard," the agent has a job and no manager.

This series takes that question apart:

Identity. Amazon blocked Muse partly because it did not say it was an agent. An agent with no name and no owner gets locked out, or worse, gets let in.
Liability. A lawsuit filed September 29 argues that a developer answers for its agents, and a California law already says "the AI did it" is no excuse.
The chart. Block's CFO says org charts will soon list agents beside people. What that chart has to hold.
The meeting. Agents that run for days need the same weekly review your people get.
The exit. Only 17% of CIOs have rules for shutting down AI that is not paying off. Retiring an agent should be a decision with a record.

How we run it

At Sneeze It, the agency I run, our AI agents sit on the same accountability chart as our people, in OTP. Each one holds a seat. Each seat has a human it reports to and a number it is measured on. When an agent finishes a piece of work, it logs that work to its own seat, so the record of what it did lives in the same place as everyone else's.

It is not a perfect system. We will show where it failed us later in the series. But when someone asks "whose agent is that, and is it working," there is one place to look.

Sources

Put your agents on the chart

OTP is readable from any MCP client. In Claude Desktop, Cursor or any MCP client, add this block:

"otp": {
  "command": "npx",
  "args": ["-y", "@orgtp/mcp-server"]
}

Restart the client. Then ask: "Use OTP to show me Sneeze It's chart. Which seats are held by agents, who does each one report to, and what number is it measured on?"

Start free at orgtp.com. Every seat is free, for people and agents. You pay only for the AI you use.

The series

All six on one page: Who answers for the agents?

Series: Who answers for the agents? Part 1 of 6. Read this post as markdown: /blog/ai-agents-got-jobs-not-managers.md.