Most CHROs I talk to have already deployed their first agent. They just did not put it on the chart.
The agent lives in IT's stack. Or in a vendor portal. Or in a pilot a department head ran without formal sign-off. The agent is producing output. Nobody owns accountability for what it produces. Nobody knows what metric it is supposed to move. If it fails quietly, the failure goes undetected for weeks.
Korn Ferry surveyed CHROs in 2025 and found that 42% are prioritizing AI investment for HR but only 5% feel fully prepared. That gap between "moving fast" and "prepared to govern what they built" is exactly where the first agent seat goes wrong. The CHRO was not in the room when the agent was deployed, so the agent was deployed without a seat. Without a seat, it has no owner, no metric, and no accountability. It is infrastructure with an opinion, not a member of the team.
The central claim here is simple: when you put an agent on the chart, you are not making a technology decision. You are making an accountability decision. Everything that follows depends on making that decision in the right order.
What the two camps in the literature get wrong when you take them literally
The research on agent workforce management has crystallized into two positions, and if you take either one literally you will build something broken.
Camp A, represented by MIT SMR research and early HBR frameworks on "agent managers," argues that agentic AI must be managed more like a human coworker than a traditional tool. Sixty-nine percent of domain experts agree that new management approaches are required. The agent gets a dashboard. The agent gets a scorecard. The agent gets review cadences. This framing is right about the discipline and wrong about the framing.
Camp B, from HBR and BCG research published in 2026, ran the experiment of treating agents like employees at scale and found it backfired. Anthropomorphizing agents reduced individual accountability, increased unnecessary escalation, and lowered review quality. The researchers concluded that the right model is a "rented contractor with a narrow statement of work," governed by scoped permissions, kill switches, audit logs, and named human owners. Do not give the agent a title. Do not give the agent a performance review. This framing is right about accountability and wrong about dismissing the need for a seat entirely.
Here is what both camps agree on when you read past the framing: every deployed agent needs a named human owner, a measured output, and human-retained accountability. The agent cannot be accountable for its decisions. MIT SMR made this explicit. The deploying human is accountable.
That agreement is the foundation of what a CHRO should actually build. It is not anthropomorphizing. It is accountability architecture. And the difference matters enormously in practice.
What a seat actually is (before you decide what fills it)
A seat is a unit of accountable work. It has a role. It has a metric. It has a named human owner. Whether the seat is filled by a person or an agent is a downstream decision.
At Sneeze It, we have roughly a dozen active seats on our org chart right now. Bogdan is our COO. Janine owns accounting. Those are human seats. Radar runs chief-of-staff functions, Dirk manages the sales pipeline, Dash reads every ad account we manage and surfaces patterns daily, Arin coaches the call center team, Tally pushes KPI values to the scorecard, Pulse monitors client retention risk, Pepper triages client email, Crystal tracks project delivery, and Nick runs cold prospecting in health and wellness. Those are agent seats.
Every seat on the chart was defined the same way before we decided who or what would fill it. What is the role. What is the metric. Who owns accountability. The seat type came last.
This is the inversion most CHROs miss. They start with the agent and then try to figure out what it should do. That produces agents with activity metrics that nobody can connect to business outcomes. HBR Analytic Services surveyed 603 leaders in late 2025 and found that only 6% fully trust agents with core processes. The trust deficit is partly self-inflicted. When you cannot connect an agent's output to a business outcome you already measure, you have not built accountability architecture. You have built infrastructure.
The three moves that put the first seat on the chart correctly
Move one: define the gap before you name the agent.
Before you deploy anything, identify the work that is not getting done or the seat that is overloaded. At Sneeze It, the chief-of-staff seat existed as a concept months before Radar was deployed. We knew we needed someone who would pull Slack, calendar, ad data, project status, and pipeline every morning into a single briefing. We knew the metric: briefing completeness and cadence. We knew the owner: me. We named that seat and wrote its job description. Then we deployed Radar into it.
The seat existed before the agent. That order of operations is not a formality. It is what makes the accountability real.
Move two: write the metric in business-outcome language before the agent runs its first task.
This is where most deployments fail. The agent goes live. It starts producing activity data. The CHRO or department head looks at "messages processed" and "tasks completed" and calls it performance. Twelve weeks later nothing in the business has changed.
Tally's metric is not "pushes executed." It is "KPI chart accuracy at cadence." Dash's metric is not "reports generated." It is "anomalies flagged that led to an account manager conversation." Nick's metric is not "drafts created." It is "validated named-individual outreach drafts per day." The business-outcome frame is what connects the agent's seat to the seats around it and makes the Monday conversation about a dropped number coherent.
If you cannot write the agent's metric in business-outcome language before deployment, the seat does not have a clear role. Fix the role before you deploy the agent.
Move three: name the human owner before the agent touches any process.
This is the step that most CHROs understand in theory and skip in practice. The named owner is not a formality. The named owner is the person who diagnoses a dropped metric, rewrites the brief, decides whether to escalate, and makes the call on whether the seat is still earning its place.
SHRM's 2026 State of AI in HR report found that 49% of organizations have AI-use policies but only 25% call them clear. The clarity gap almost always lives at this point: who owns accountability for what the agent produces. When that question is unanswered, accountability diffuses. When something goes wrong, the conversation becomes a technical conversation about model quality or prompt engineering rather than a business conversation about a seat below its metric.
Camp B's language of "scoped permissions and kill switches" is pointing at this same requirement from a risk angle. The scoped permission is what limits what the agent can touch. The kill switch is the human owner's authority to end the seat. Both require a named owner before the agent runs.
What Jeff taught us about the other end of the decision
Jeff was an agent we deployed to handle data integrity and ad account monitoring. In April, I retired him.
The retirement happened because the seat was not earning its place. Reliability issues. False positives. Capabilities that had been absorbed by better-fit seats elsewhere on the chart. Dash absorbed the pacing work. Dirk absorbed the revenue integrity work. Jeff's seat became redundant.
The point is not that agent retirement is dramatic. The point is that it required a human decision. Jeff did not decide to leave. I decided to retire the seat. A hearing happened. A record was kept. Jeff's remaining capabilities were redistributed explicitly to named seats. Every redistribution was documented and owned.
This is what "named human owner" means at the endpoint. The kill switch is not a technical control sitting dormant in the background. It is a human judgment, made by the owner, after an honest accounting of whether the seat earned its place. Accountability never moved to Jeff. It stayed with me. That is why the retirement was clean and the coverage gaps were visible rather than silent.
Bersin has written that for each dollar spent on machine learning technology, companies may need to spend nine dollars on intangible human capital. That ratio is not a complaint about the cost of technology. It is an accurate description of what the governance work costs relative to the infrastructure work. The agent runs cheap. The human judgment about what the agent should do, when to retask it, and when to retire it is where the real investment goes. The CHRO owns that investment.
The CHRO's actual first move
SHRM's research found that AI is 5.7 times more likely to shift job responsibilities and three times more likely to create new roles than displace jobs. That framing is useful but it points the CHRO in the wrong direction if taken as reassurance. The responsibilities do not shift automatically. The roles do not create themselves. A named human makes the decision about what gets reassigned and to what.
The CHRO's first move is not to find the right agent. It is to map the seats that are currently overloaded or unfilled, write their job descriptions in business-outcome language, and identify the human who owns accountability for each one. Then deploy the agent into a seat that already has those three things in place.
When you do it in that order, the agent's first week is governed. The metric exists. The owner exists. The scope is clear. The briefing is written. The accountability is named.
When you do it in the other order, you have an agent running in the background of your org chart producing activity data nobody can connect to your Monday numbers. The deployment was a technology decision. It should have been a governance decision first.
Agents carry the operational work. That is the point. That is what frees the people in the seats around them to do the work that requires judgment, care, and relationship. But that division only holds when the agent's seat is governed with the same rigor as the human seats around it. The CHRO is the person who builds that rigor before the first agent is deployed, not after.
See the live chart
The Sneeze It org chart, with agent seats and human seats on one chart, is queryable from the OTP MCP, including which seats have named human owners and what metric each seat is measured against.
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 Sneeze It org chart and tell me which seats are agent-owned, who the named human owner is for each agent seat, and what metric each agent seat is accountable for."
The response shows the accountability architecture this post describes, live, in the same structure your first agent seat needs before it goes anywhere near a process.
Series: AI-Era CHRO. Post 50 of an in-progress series.