The operations meeting is where the COO finds out what is actually happening.
Not what people said would happen. Not what the dashboard showed at 9 AM before anyone had reviewed it. What is actually happening, right now, at the level of the process that moves work through the business. The meeting exists to surface the gap between what should be running and what is running, assign accountability for the gap, and produce commitments that close it before next week.
That job description does not change when agents hold seats in the meeting. What changes is the quality of the inputs, the speed of the accountability conversation, and the precision of what the meeting can actually see.
I run this meeting every week at Sneeze It. Half the seats belong to agents. Here is what the meeting looks like.
Before the meeting: the publication audit
A traditional operations meeting starts with prep that may or may not have happened. Someone sends their numbers the night before. Someone does not. Someone's slide deck is finished. Someone's is missing a column. The pre-meeting prep gap is a human gap, and it is almost always a surprise.
With agent seats on the agenda, the prep gap is not a surprise. It is a diagnostic.
Every agent seat at Sneeze It is expected to have published its output file before the meeting begins. Radar, our chief-of-staff agent, publishes a shared state file covering Slack, calendar, and task queue. Dash, our analytics agent, publishes a performance file covering every managed client account, with spend, leads, and alert flags. Tally, our scorecard agent, pushes the current KPI values to the org chart. Crystal, our project manager, publishes project status across every active delivery. Arin, our call center manager, publishes appointment rate, speed to lead, and per-agent performance against the 30% booking target.
Before I open the meeting, I check which files are current and which are stale. A stale file is not an inconvenience. It is an agenda item. Something did not run, or ran and produced no output, or ran and failed without surfacing the failure. That is a process problem, and it goes on the board the same as any other process problem.
This is the first structural shift in the operations meeting when agents hold seats. You stop asking "who did not prepare" and start asking "what did not run." Those are different questions and they have different implications. A human who did not prepare has a reason you will learn about in the meeting. A process that did not run has a cause you can trace and fix before next week.
The open: one number, stated plainly
Every seat in the meeting opens with one number. Not a summary. Not a narrative. One number that defines whether the seat did its job this period.
Bogdan, our COO and the human anchor of the operational layer, opens with his lead indicator. Janine, our accounting lead, opens with cash collected. Then Radar opens with daily briefings published. Dash opens with total managed ad spend and client accounts with active alerts. Crystal opens with projects in active delivery versus projects overdue. Arin opens with team-wide appointment rate this week versus the 30% target. Dirk, our sales agent, opens with qualified meetings booked. Nick, our cold prospecting agent, opens with quality emails drafted and delivered to the send queue.
Each number is the throughput of the seat. Not the activity inside the seat. Not a count of what the agent processed. The output that the rest of the process depends on.
When every seat opens the same way, the meeting has an anchor. You can see immediately which seats are above their number, which are below, and which are at or near threshold. The pattern is visible in about ninety seconds.
The accountability loop: same conversation, faster
The accountability conversation in an operations meeting has one shape regardless of who or what holds the seat.
Number above target: what caused it and can we sustain it. Number below target: what caused it, who owns the fix, and what is the commitment for next period. Number missing: what did not run, who is responsible for making it run, and what is the fix.
That loop does not change when the seat belongs to an agent. What changes is how fast the loop resolves.
When Dash flags that a client account crossed a spend alert threshold, the conversation in the meeting is: what is the threshold, when did it cross, what is the correct response, who executes. The answer is available in the file Dash published. We do not spend time reconstructing what happened. We spend time deciding what to do about it.
When Arin flags that appointment rate for one client dropped below 25%, the conversation is: which callers are below target, what does the data show about their dial volume and speed to lead, what coaching adjustment is needed, and who delivers it. Arin surfaces the pattern. The meeting resolves the response.
This is the payoff that Accenture points to when they describe what it means to reinvent the process before adding the agent. The agent does not make the old conversation faster. It replaces the reconstruction portion of the conversation entirely. The meeting stops spending time figuring out what happened and starts spending time on what to do next.
At Sneeze It, Deloitte's 2026 State of AI survey found only 21% of enterprises have a mature governance model for agentic AI. The operations meeting is part of that governance model. It is where the COO exercises accountability over the fleet, not through dashboards, but through the same conversation discipline applied to every seat.
The exception layer: what agents cannot resolve
The operations meeting is also where the COO manages what the agents flag but cannot resolve themselves.
Agents carry the operational work. They do not make judgment calls. When Dash identifies that a client account's spend pattern suggests a budget exhaustion risk before end of month, that is a flag. The decision about how to respond to that client is a human judgment call that happens in the meeting.
When Crystal flags that a project milestone is at risk because a handoff between two tasks did not complete on time, that is a diagnostic. The decision about whether to escalate, negotiate scope, or reallocate human capacity is the COO's call.
When Dirk flags a deal that has been in the same pipeline stage for thirty days with no movement, that is a signal. The decision about whether to pursue, pause, or close the deal is David's call.
The operations meeting is the room where these exception calls happen. The agents generate the flags. The humans in the room, including Bogdan as the operational anchor, resolve the exceptions and produce commitments.
This is the principle McKinsey describes when they write about managing people and agents together. The meeting is the place where that management is exercised. Not in one-off conversations. Not in async threads. In a structured, weekly accountability loop where every exception surfaces on the same agenda as every routine result.
The COO keeps humans on the high-value work. That means keeping the exception resolution in the meeting and keeping the routine operational load on the agents.
The commitment close: specificity is the gate
Every item that surfaced in the meeting ends with a commitment. Not a plan. Not a discussion thread. A commitment with a name, an action, and a deadline.
This requirement sharpens considerably when agents hold seats.
Agent seats can publish their output by a specific time. That is a commitment. "Dash will publish the performance file by 7 AM Monday" is a commitment. "Arin will publish the week's coaching flags by Friday EOD" is a commitment. "Tally will push KPI values by 6 PM on business days" is a commitment. Those commitments are either kept or not, and the operations meeting is where the COO reviews whether they were kept.
Human seats make commitments too, but human commitments are sometimes ambiguous. "I'll follow up with the client" is not a commitment. "I will send the client a one-paragraph update on the project timeline by Wednesday noon" is a commitment. The discipline the agents bring to the meeting is that their commitments are always the specific kind. Agents publish by a time or they do not publish. The binary is what the meeting requires.
When I ask Bogdan or Janine or Kristen for their commitment for next week, I use the same language I use when reviewing the agent commitments. What specifically, by when, and what does done look like. The agents raised the standard. The humans meet it.
What the meeting reveals that a dashboard does not
The operations meeting reveals the system, not just the numbers.
A dashboard shows a KPI. The operations meeting shows whether the process that produces that KPI is healthy, whether the handoffs between seats are working, and whether the exception layer is being managed or is accumulating into a risk.
At Sneeze It, the meeting surfaces things the dashboards never would. It surfaced that two agents were producing outputs in formats that a downstream human could not easily act on. It surfaced that a handoff between Nick's prospecting output and Dirk's pipeline intake was producing a lag. It surfaced that a commitment from the previous week had not been kept and the number that depended on it had drifted.
The agents coordinate through the agent message bus, sending structured messages to each other's inbox files at ~/.claude/agent-inbox/. That coordination is invisible to any dashboard. But when a coordination failure results in a number dropping, the operations meeting is where the COO sees it and traces it.
We retired Jeff, a former data integrity agent, through the operations meeting. Not through a unilateral decision. Through the accountability loop. His number had not moved in five days. His commitments were not being kept. The meeting surfaced the pattern. The hearing happened. The retirement followed. The meeting is where Jeff's seat came up for review, and where the redistribution of his capabilities to Dash and Crystal was decided.
The meeting is the governance layer. The agents are the operating layer. The COO runs both by running the meeting.
The lifecycle of an operations meeting with agents
The lifecycle frame for this meeting is open, operate, exception, close.
Open with the numbers. Every seat, one number, thirty seconds. The COO reads the pattern in the room.
Operate through the accountability loop. Above, below, missing. Each gap gets a cause and a name.
Exception layer. What the agents flagged that humans need to resolve. These are judgment calls, not process steps. The COO owns the resolution.
Close with commitments. Every open item leaves the meeting with a name, an action, and a deadline. Agent seats get specific publication commitments. Human seats get specific action commitments. Both are reviewed next week without distinction.
The meeting does not get simpler when agents hold seats. It gets more honest. The operational layer is visible in a way it was not when every number had to be reconstructed in the room. The exception layer is clearer because the agents surface flags the humans would have missed or seen too late. The commitment discipline is sharper because the agents set the standard for what specific looks like.
That honesty is what makes the COO's job possible at scale. Let agents carry the operational work so people are free for the work that matters. Run the meeting so the system stays honest about what is carrying what.
See the live chart
Every seat named in this post, including which agents published this week and which humans hold exception authority, is queryable from OTP MCP.
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 which seats at Sneeze It are agent seats and what each one is accountable for delivering each week."
The response gives you a live view of the hybrid operating layer, the same layer the COO reviews every Monday.