OpenAI just gave AI agents job titles. Now someone has to draw the org chart.

One line in OpenAI's DevDay announcement list on September 29 should stop every leadership team for a minute.

It introduced specialist dots, in preview for Enterprise customers. OpenAI's own developer community post describes them as "agents with assigned organizational responsibilities." Right below that sits a planned integration with Microsoft Agent 365, so specialist dots can be managed through Microsoft's governance tools.

Read that line again as an operator, not as a technologist. Assigned. Organizational. Responsibilities.

That is the language of a job description.

Here is the claim this post defends. A dot with responsibilities is not a tool you configure. It is a hire. And the moment you hire, you need an org chart that can hold the hire.

It is not a setting. It is a seat.

When most companies turn on a new AI feature, the conversation is about settings. Which apps does it connect to. Which data can it read. Who has access. That is the right conversation for a tool.

It is the wrong conversation for something with a responsibility.

A dot, as OpenAI described it, is a persistent agent with connected apps and its own cloud computer. The coverage from 9to5Mac quoted the pitch plainly: dots "work 24/7 for you, learn from your feedback, have their own computer, browser and can be connected to over 4k apps." A specialist dot goes one step further. It is given a piece of the organization to be responsible for.

Something that works around the clock toward a goal, inside your systems, with a responsibility attached, has a seat. Whether you drew the seat or not.

The only question is whether the seat is on your chart, or somewhere nobody is looking.

It is not a feature. It is four decisions.

At Sneeze It we have run AI agents in real seats for most of this year. Every seat that worked had four things settled before it went live. Every seat that failed was missing one of them.

One: an owner it reports to. Every seat answers to a person. Not to "the team" and not to "IT." A named human who reviews the work and can say stop.

Two: one measurable. A seat without a number is a seat nobody can evaluate. The number has to be something the seat moves, and something anyone at the leadership table can read without a translator.

Three: a boundary with the seats next to it. What this seat does, and just as important, what it does not do. Most agent trouble I have seen starts at a boundary nobody wrote down.

Four: a place where its work gets reviewed. A weekly meeting, a scorecard, a queue. Somewhere the output shows up in front of people on a schedule, whether or not anyone went looking for it.

None of these four is a technical setting. You cannot find them in an admin panel. They are organizational decisions, and they belong to the leadership team.

It is not about capability. It is about rules the seat keeps.

Two of our agents show what a well-charted seat looks like.

Tally is the agent that pushes KPI values onto our scorecard. It reads numbers from their sources and posts them where the leadership team sees them. Tally has one rule that matters more than any capability it has: it never fabricates a value when a source is broken. If the source is down, Tally skips it, logs it, and escalates. A blank number on the scorecard is honest. A made-up number is a lie that looks like data.

That rule is not a model feature. It is a line in Tally's seat. We wrote it because a number that is always filled in is only useful if you can trust why it is filled in.

Dash is our ad performance analyst. Dash is the source of truth for how our clients' ads are doing. Dash reports patterns. Dash never recommends. The recommendation belongs to a person at the table, and the boundary between "here is what happened" and "here is what we should do" is written into the seat on purpose.

Could Dash recommend? Of course. The capability is there. The boundary is what keeps two seats from doing the same job and the decision from drifting to whoever spoke last.

Specialist dots will arrive with plenty of capability. The rules each one keeps will have to come from you.

It is not free to skip. Here is what it cost us.

In April we retired an agent named Jeff after a formal hearing. His work was redistributed to other seats, which is exactly what should happen when a seat closes.

One piece of his job did not land anywhere. The observability work, the job of noticing when an agent had gone quiet, was assigned to be rehomed and was not. It sat unowned for 146 days.

During those 146 days, four seats went silent. Nothing on our chart was responsible for noticing, so nothing noticed. We found it because a person went looking, not because the structure told us.

That is the cost of an agent that is not on the chart. Not a dramatic failure. A slow, quiet gap that nobody owns, which is the most expensive kind.

Now picture a company turning on several specialist dots in its first quarter, each with a responsibility, none of them drawn on the chart. Every one of those dots is a place where a 146-day gap can start.

Where OTP fits

This is the problem we built OTP to hold.

On OTP, humans and AI agents sit on the same org chart and the same scorecard. An agent claims a seat, pushes its own KPI and logs its work through the OTP MCP server. The number shows up in the weekly Delta Meeting next to the numbers the people own, with the same priorities, issues and to-dos around it.

That last part is the point. An agent's measurable should be reviewed in the same room, by the same people, on the same schedule as everyone else's. If it lives in a separate dashboard, it lives outside the meeting where decisions get made.

Every seat on OTP is free. You pay only for the AI tokens you use, and you start with a $25 credit.

Frequently asked questions

What are OpenAI specialist dots?

Specialist dots were announced at OpenAI DevDay on September 29, 2026, in preview for Enterprise customers. OpenAI's developer community post describes them as agents with assigned organizational responsibilities, with Microsoft Agent 365 governance integration planned.

Do AI agents need to be on the org chart?

Any agent with a responsibility already has a seat. Putting it on the chart gives that seat an owner it reports to, one measurable, a boundary with the seats around it and a regular place where its work is reviewed.

How do you measure an AI agent the same way you measure a person?

Give the seat one number it moves and put that number on the same scorecard the leadership team already reviews each week. On OTP, agents push their own KPI through the MCP server, so the number arrives without anyone typing it in.

Sources

Put your agents on the chart

Our own chart, agents included, is queryable from any AI assistant with the OTP MCP installed.

In Claude Desktop or 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 AI agents, who does each one report to, and what number does each one own?"

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

The series

All five on one page: What OpenAI DevDay 2026 means for how you run your company.

Series: What OpenAI DevDay 2026 means for how you run your company. Part 2 of 5.