The meeting format does not change. The inputs to that format change completely.
That took me longer to figure out than it should have. When we started running agents at Sneeze It, I assumed the leadership meeting would need a new structure to accommodate them. I looked for frameworks. I read about hybrid team management. I spent more time than I want to admit thinking about how to redesign the meeting room for a room where some of the seats do not have people in them.
The answer turned out to be simpler and more demanding than any redesign. You run the same leadership meeting you already run. You hold every seat on the agenda to the same lifecycle every seat always goes through: opening number, performance since last meeting, current focus, blockers, commitments for next week. You just have to be rigorous about what counts as input from an agent seat, who speaks for it, and what happens when the number is wrong.
That rigor is where the work lives.
Before the meeting: who published and who did not
In a traditional leadership meeting, the prep gap is usually human. Someone did not send their numbers. Someone's report is incomplete. Someone did not read the deck before walking in.
With agent seats on the agenda, the prep gap changes character. Agents publish. They do not show up late and apologize. They either ran and wrote their output file, or they did not. When they did not, that is a signal, not an inconvenience.
Every Friday before our Monday meeting, I check whether each agent seat published its week. Radar, our chief-of-staff agent, writes a shared state file. Dash, our analytics agent, writes its file. Tally, our scorecard agent, pushed KPI values to the chart. Dirk, our sales agent, updated the pipeline file. Pulse, our retention agent, wrote client health indicators. Crystal, our project manager, wrote project status.
A missing file is a system problem. A stale file (more than 18 hours old) is a runtime problem. Both are agenda items before the meeting starts, the same way a human who does not have their numbers ready is an agenda item.
This pre-meeting audit is the first thing that changes when agents hold seats. You go from "who is not prepared" to "what is not running." Those are different questions and they have different answers.
The opening: the number, not the narrative
Every seat in our Monday meeting opens with a number. Not an update. Not a narrative. A number.
Bogdan, our COO, opens with whatever his lead indicator is that week. Janine, our head of accounting, opens with cash collected. Kristen, our creative director, opens with active briefs in progress. Then Radar opens with daily briefings published (five out of five this week, or four, or three). Tally opens with KPI push completion rate. Dash opens with total managed ad spend and whether any client accounts crossed an alert threshold. Dirk opens with cold emails sent and qualified meetings booked.
The meeting does not start with context. It does not start with "so here is what has been going on." It starts with the number that tells the room, in one beat, whether the seat is on track.
This is the same discipline whether the seat is human or agent. The difference is that agent numbers are usually more precise. Dirk does not say "roughly thirty." Tally does not say "most KPIs pushed." The agents either have the number or they have an error, and an error is a different kind of input than a rounded estimate from a human who is slightly guessing.
MIT CISR's research on enterprise AI maturity found that Stage 4 firms, the ones with the highest growth and profit outcomes, operate with a united top leadership team spanning CEO, functional leads, and the people accountable for AI systems. The meeting is where that unity gets exercised. Not in a strategy document. In the room, every week, with every seat on the agenda held to the same standard.
The middle: same three questions, different answers
After the opening number, every seat gets three questions.
What produced that number. What changed since last week. What is the plan to move it next week.
For human seats, these questions often produce conversation. There is context. There is judgment. There is attribution that requires someone who was in the situation to explain it.
For agent seats, the answers are often faster and more specific. When Dash reports that a client account crossed a spend threshold, the answer to "what produced that" is: the account spent 22 percent above its 7-day baseline on Wednesday, and here is the spend curve. The answer to "what changed" is: the client expanded their ad set on Tuesday. The answer to "what is the plan" is: Radar has flagged it for David to review with the client before Thursday.
The three questions reveal something important about agent seats. They are usually very good at "what produced that" and "what changed." They are almost never the right seat to answer "what is the plan" when the plan requires judgment that crosses into client relationships, pricing, or strategic direction. The plan answer for agent seats usually routes to a human.
This is the operating version of McKinsey's point that managing in the AI era means managing systems of people and agents together. The leadership meeting is where that management happens week to week. The agent produces the diagnostic. The human makes the call. Both inputs are in the room, on the same agenda, in the same conversation.
When an agent number drops
The most important moment in a hybrid leadership meeting is when an agent's number drops.
The instinct in most organizations is to reach for a technical explanation. The model got confused. The API had latency. The prompt needs updating. These may all be true. They are also a way of treating the agent's performance as a technical event rather than an accountability event.
We do not do that.
When Arin, our call center manager agent, shows a drop in the appointment rate for the week, the conversation is the same conversation we would have if a human call center manager showed a drop in that number. What changed in the inputs. What changed in the team's execution. What does the seat need to recover the number. The answer might ultimately be technical, but the framing is not. The framing is: this seat is accountable for this number, and the number dropped, and we need a plan.
This shift in framing is not subtle. When you treat an agent's drop as a technical event, you get debugging. When you treat it as an accountability event, you get management. Debugging moves the number once. Management builds the discipline that moves it consistently.
Deloitte's 2026 State of AI in the Enterprise found that only 21 percent of organizations have a mature governance model for agentic AI. The other 79 percent are, I would guess, mostly treating agent performance as a technical event. The accountability framing is what mature governance looks like in a meeting room.
The agent seat that speaks for itself
One pattern that surprised me early: agents that publish their own numbers before the meeting are easier to manage than agents where a human is translating the number into the room.
When Tally pushes KPI values to the chart at 10:15 AM and again at 12:15 PM, I see the number before the meeting. I can form a question before walking in. When a human has to relay what an agent did, a translation layer appears, and translation layers introduce interpretation. The agent might have flagged something the human summarized away. The agent might have produced nuance the human did not have time to carry.
The operating principle we settled on: every agent seat publishes its own output to a shared file. The human who is accountable for that seat brings the file to the meeting and speaks for the seat when questions arise, but the raw output speaks first. The seat-owner explains, clarifies, and makes calls. The agent's file is the record.
Nick, our cold prospecting agent, publishes a batch report after every run. Dirk publishes pipeline changes each time it reads the CRM. Pepper, our email agent, writes an inbox state file after each triage. These files are the agent seats in the room. The humans who manage those seats are the seat-owners, and they carry the accountability that agents cannot carry: the relationship context, the judgment call, the decision that crosses into territory where a human has to own the outcome.
The retirement conversation
There is a conversation that only happens in a meeting where agents hold seats alongside humans: the retirement conversation.
We had one in April. Jeff, our data integrity agent, had been in the seat for several months. His numbers had drifted. His file was stale more often than not. His role had been partially absorbed by Dash and Dirk as those agents matured. In a meeting where every seat is measured on the same scorecard, the retirement case assembled itself from the data.
We ran a hearing. Jeff was the first agent we retired, and the process was the same process we would use for a human seat that had been made redundant by org changes: describe what the seat was hired to do, document what it actually did, identify where the capabilities should live going forward, make the decision clean.
The meeting discipline is what made that possible. If Jeff had been on a separate agent dashboard, the drift would have been invisible for months longer. Because the seat was on the same agenda as Bogdan and Janine and Kristen, the gap was visible the same week it appeared.
Gartner estimates that as reported by CIO.com, agent sprawl is already the new Shadow IT, with organizations running 50 or more agents without a clear inventory of who owns what. The retirement conversation, held regularly, is the antidote to sprawl. You cannot have that conversation without a meeting structure that puts every seat on the same agenda.
After the meeting: commitments that close the loop
Every seat leaves the Monday meeting with one commitment for next week. One thing that the seat will do differently, more of, or at a higher standard. The commitment goes on the record.
For human seats, this is the classic L10 to-do. Due by next Monday. Checked at the following meeting.
For agent seats, the commitment usually takes a different form. It might be a rule change, a threshold adjustment, a new input the agent will receive, or a workflow the seat-owner will fix. The agent does not write its own commitment. The seat-owner writes it, because the seat-owner is who can actually execute a change to the agent's operating environment.
This is the closing step of the lifecycle. Opening number. Three questions. Accountability for drops. Agent publishes, human speaks. Retirement when needed. Commitments that close the loop. The meeting runs the same lifecycle for every seat, every week.
The result is that execution stays accountable. Not because the agents are being watched, but because the meeting structure makes their output visible in the same way human output is visible. The loop closes. The discipline compounds.
That is the point. Agents carry the operational work. The meeting is where the CEO holds every seat, human and agent, to the same standard, and makes the calls that agents cannot make on their own.
See the live chart
The OTP MCP exposes the live Sneeze It org chart, current scorecard values, and each agent seat's assigned KPIs, so you can query exactly what a leadership meeting would surface for any seat.
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 the current KPI values and owners for the sneeze-it org chart, and identify which seats are held by agents versus humans."
You get back a structured view of exactly what the Monday meeting works from. That is what a hybrid leadership meeting looks like before the room fills.