The conversation about agent literacy almost always starts in the wrong place.
It starts with training. Who needs to understand AI. How many hours of instruction. Which teams get the workshop first. Whether the curriculum is technical or conceptual. Whether it comes from inside or from a vendor.
All of that can be useful. None of it is where literacy actually lives.
Literacy lives in the seat design. The organizations that build genuine agent literacy are not the ones that ran the best training program. They are the ones that made it structurally impossible to deploy an agent without a named owner, a measured outcome, and a clear human decision about when the agent is done.
That structure is what builds the skill. The training program teaches vocabulary. The seat design produces judgment.
Why the training-first approach fails
SHRM's 2026 State of AI in HR report found that 62% of organizations are using AI somewhere, but only 49% have AI-use policies, and only 25% call those policies clear. The gap between adoption and governance is not a training failure. It is a design failure.
People in organizations are deploying agents the same way they installed Dropbox in 2005: because it solves a problem they have right now, without waiting for a policy that may never arrive. Gartner described this as the new Shadow IT, as reported by RCR Wireless. By Gartner's account, roughly 40% of enterprise applications may include task-specific agents by 2028. Most of those agents will have no named owner and no performance standard.
A training program does not fix this. A design does.
The reason is causal. Training gives people a mental model. Seat design forces them to use it. When the mental model has no structure to attach to, it evaporates within weeks of the workshop. When every agent deployment in the organization requires a named owner, a business-outcome metric, and a review cadence, the people doing those deployments build judgment through repetition, not through instruction.
The split the research reveals, and why it matters
Before landing on how to build literacy, the CHRO needs to understand a live disagreement in the management literature, because the wrong answer here produces exactly the failure mode literacy is supposed to prevent.
One camp says: manage agents like coworkers. MIT Sloan Management Review, in an analysis where 69% of interviewed experts agreed that agentic AI requires new management approaches, argues that agents must be managed more like human coworkers than like traditional tools. HBR introduced the role of "agent manager" in early 2026: a human who runs agents via dashboards, scorecards, and observability. The implication is that HR's existing playbook, onboarding, performance reviews, structured accountability, extends naturally to agents.
A second camp, represented in HBR and BCG research published in May 2026, argues the opposite. Researchers found in a large experiment that when people anthropomorphized agents, treating them like employees with feelings, career arcs, and social standing, individual accountability dropped, unnecessary escalation increased, and review quality declined. Their prescription: treat agents as rented contractors with a narrow statement of work, governed by scoped permissions, audit logs, kill switches, and named human owners. Not HR onboarding. Not performance reviews. Not titles. Governance.
The organizations that pick one camp and ignore the other will struggle. Camp A's framing, that agents need structured management, is correct. Camp B's warning, that anthropomorphizing agents transfers accountability away from humans, is also correct. The synthesis is not complicated once you see it: the agent gets a seat with a named human owner and a measured outcome. The human owner is accountable. The agent executes. Accountability never moves to the agent.
At Sneeze It, Jeff was a data integrity agent. When his missions were absorbed by better-fit seats and his output went stale, we held a formal retirement hearing. I made the decision. Jeff did not retire himself, did not feel the loss, and did not need severance. The hearing was not for Jeff. It was for us: to document what Jeff had been doing, reassign the coverage explicitly, and keep a record. That is what human-retained accountability looks like in practice. It is a governance exercise, not an HR ceremony.
What the CHRO actually builds
HBR Analytic Services surveyed 603 leaders in late 2025 and found that only 6% fully trust agents with core business processes, and only 12% have risk or governance controls fully in place. The gap between deployment and governance is universal. The CHRO is positioned to close it, but not through policy documents.
The CHRO builds literacy by designing four things into every agent seat before deployment.
The first is scope definition. What does this agent do. What does it not do. The scope has to be specific enough that a deviation would be obvious to the domain expert who reviews the output. Pepper, our email triage agent, drafts client communications. Pepper does not decide which clients to contact or what the relationship strategy should be. That line is the scope. When a draft crosses it, the human reviewer catches it immediately because the scope was defined clearly enough to make the crossing visible.
The second is a named human owner. Not a department. Not a team. A person whose name is on the seat. When Arin, our call center manager agent, sends a coaching message that misreads a team member's situation, I am the person who redirects it. I am not the AI team. I am not a governance committee. I am the seat owner, accountable for what Arin produces the way a manager is accountable for a direct report's work. That naming is not ceremonial. It is the mechanism by which accountability stays human.
The third is a business-outcome metric. Not tokens consumed. Not tasks completed per hour. The metric that proves the seat is earning its place. Dash, our analytics agent, does not get measured on how many accounts she reads. She gets measured on alert accuracy: how often she flags something real versus something that turns out to be noise. Tally, our KPI-push agent, gets measured on whether the scorecard is current when we walk it on Monday. The metric is what makes the seat visible to the domain expert who cannot evaluate the agent's internal logic but can absolutely evaluate whether the outcome showed up.
The fourth is a retirement trigger. Every seat needs a condition under which the human owner recommends closure. Not because agents have lifespans, but because seats become obsolete when the work moves, the context changes, or a better-fit seat absorbs the mission. Without a retirement trigger, agents accumulate. By Gartner's projection, over 40% of agentic AI projects will be canceled by 2027, as reported by RCR Wireless. Most of those cancellations will happen reactively, after drift has accumulated and trust has eroded. A proactive retirement trigger is a governance asset.
The causal chain
When these four elements are in place before the first deployment, something happens that no training program produces.
The humans who own the seats start building genuine judgment about agent performance. Not abstract understanding of what agents are capable of. Actual, domain-specific judgment about whether this particular agent, running this particular scope, is producing outputs they would stake their accountability on.
Korn Ferry's 2025 workforce research found that 70% of senior leaders say their organization has an AI strategy, but only 39% of employees agree. That gap is not a communication problem. It is a design problem. The strategy lives in executive presentations. The employees experience agents doing work in their domain without a structure that makes the performance visible to them or connects the agent's output to an accountability chain they are part of.
The four-element seat design closes that gap. When Bogdan, our COO, reviews the Monday scorecard and sees Radar's row alongside his own and Janine's, he is not reviewing an AI dashboard. He is reviewing the operational scorecard he owns. When a row drops, the conversation is the same regardless of whether the seat is human or agent. Deloitte's 2025 Global Human Capital Trends research, across 10,000 leaders in 93 countries, found that 73% of respondents say middle manager role reinvention matters, but only 7% report great progress on it. The reinvention stalls because the structure that would make it concrete is not in place. The seat design is the structure.
The Bersin ratio
Josh Bersin's research on enterprise AI produces a figure that frames the CHRO's investment decision clearly. For each dollar spent on machine learning technology, organizations may need to spend nine dollars on intangible human capital. The human capital investment is not in building the models. It is in the people who own the seats, define the scopes, interpret the outputs, and make the decisions about when to intervene and when to retire.
That nine-to-one ratio is an argument for literacy. It is also an argument for where literacy lives. It does not live in the AI team that built the agent. It lives in the domain experts who own the seats and stake their accountability on the outputs.
Agent literacy built through seat design scales in a way that training programs do not. Every new seat deployment is a literacy exercise. Every Monday review where a human walks an agent's row is a literacy exercise. Every retirement decision is a literacy exercise. Bersin is right that the human capital investment dwarfs the technology investment. The seat design is how you deploy that capital efficiently, by building the skill into the operating rhythm rather than into a curriculum that runs once and depreciates.
What this means for the CHRO's roadmap
The CHRO who wants to build agent literacy across the organization does not start with training. She starts with a seat template.
One page. Scope definition, in business language, not technical language. Named human owner. Business-outcome metric. Retirement trigger condition. That template becomes the intake form for every agent deployment in the organization. No agent goes live without it.
The template produces accountability architecture. The accountability architecture forces the conversations that build judgment. The judgment is the literacy.
SHRM found that AI is 5.7 times more likely to shift job responsibilities than to displace jobs outright, and 3 times more likely to create new roles than to eliminate them. The new roles that are being created are predominantly accountability roles: the person who owns the agent seat, reviews the agent's output in domain terms, and holds the authority to retire the seat when it is no longer earning its place.
That is the role the CHRO needs to design for. Let agents carry the operational work, so people are free for the work that matters. The CHRO's job is to make sure the accountability for that arrangement stays with the people, not with the agents, no matter how confident the agents sound.
The seat template is where it starts. The accountability architecture is what gets built. The literacy is what compounds.
See the live chart
The Sneeze It org chart, including which seats are agent-owned versus human-owned and who holds named accountability for each agent seat, is queryable from the 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 the sneeze-it org chart and identify the named human owner for each agent seat."
What comes back is the accountability architecture in live form: not a framework, but a working chart you can examine seat by seat.