The Agent Is Not in the Org Chart
Enterprise agents need a workforce record, not just an identity.
Enterprise agent governance is moving from access control to recordability.
The next control question is not only whether the agent has a credential. It is whether the organization can prove who the agent was, who owned it, what work it performed, which authority applied, and where the evidence survives when the action is reviewed later.
This distinction matters because enterprise systems were built around different record types. HR systems record employees, jobs, managers, cases, and transactions. Identity systems record credentials and access. Workflow systems record cases, tasks, triggers, and approvals. Security systems record events. GRC systems record controls and exceptions.
An agent may sit between all of them.
A recruiting agent may influence a candidate workflow. A learning agent may recommend or assign activity. A case-routing agent may move an employee issue. A workflow agent may act through a vendor platform or internal integration. Each system may capture part of the event. None may preserve the whole actor record.
Brief 005 described the authority gap. Part II describes the recordability gap.
If the enterprise cannot represent the agent as an accountable non-human participant, it will struggle to govern the agent’s authority, audit its work, or retire its role when the business context changes.
- Identity is necessary but insufficient. AI agents increasingly require inventory, ownership, visibility, and access governance. But an identity record is not the same as a workforce record. Identity can show that a non-human actor exists. It does not automatically connect that actor to business role, human owner, work context, approved purpose, action history, and audit trail.
- HR and workforce systems expose the gap. Many enterprise workforce systems were designed around people, roles, jobs, managers, cases, and transactions. They were not naturally designed to record, manage, or audit non-human workflow actors as durable participants in the workforce system of record.
- Cross-system traceability is becoming the proof burden. Agent work can leave fragments across HRIS, LMS, IAM, workflow, GRC, SIEM, and vendor systems. Isolated logs may show the transaction. They may not show the accountable non-human actor behind the work, the human authority behind the actor, or the evidence trail that survives review.
- Offboarding is a consequence, not the spine. If the agent is not represented as a workforce actor, the enterprise will struggle to revoke, transfer, narrow, or audit its authority when the owner, task, workflow, or business context changes. That reinforces Brief 005 without repeating it.
- Learning systems are the next signal surface. The evidence does not yet support a broad claim that learning records are already polluted by agents. But the control question is clear: when agents participate in learning or workforce workflows, can the system distinguish the learner, manager, application, vendor, and non-human actor in the record?
From authority gap to recordability gap.
Brief 005 focused on what happens after an agent receives authority.
Part II asks a prior question: where is the agent recorded as an accountable participant in the work?
The answer is often split. The credential may sit in identity infrastructure. The workflow execution may sit in an automation platform. The business action may sit in HRIS, LMS, recruiting, case management, or knowledge systems. The risk exception may sit in GRC. The security signal may sit in SIEM. The vendor may hold its own activity log.
Each record can be accurate and still leave the organization unable to answer the governing question: which non-human actor performed or influenced this work, under whose authority, and with what surviving audit trail?
Identity does not equal a workforce record.
Identity governance can inventory agents, assign owners, scope credentials, and enforce access rules. Those controls matter. They are the foundation for any serious agent program.
But the org-chart thesis is not an anti-IAM thesis.
The issue is that workforce accountability requires more than an account object. A durable workforce record connects identity to business role, human owner, work context, approved purpose, action history, authority changes, and revocation evidence.
If an agent can act in HR, learning, workforce service, knowledge, or approval flows, the enterprise needs a way to distinguish the employee, the manager, the system, the vendor, and the non-human actor. Otherwise, the enterprise can see that work happened while losing the actor model needed to govern it.
The transaction can survive while the actor disappears.
The strongest HRIS point is narrow: a system may preserve the business transaction without preserving the non-human actor as a durable participant in the workforce record.
That matters because workforce processes are accountability systems, not only data stores.
A recruiting action, employee case route, learning assignment, time-off step, skills update, or performance flag does not only need a timestamp. It needs provenance: who or what caused it, who was accountable, what authority applied, and what record survives if the context is challenged later.
Agent-mediated work strains that model.
The human employee may appear in the business record. The approving manager may appear. The application may appear. The agent may be treated as a tool, integration, or background workflow rather than as the actor whose work requires review.
Cross-system traceability becomes the operating test.
The org-chart problem gets harder when the agent crosses systems.
Enterprise agent governance requires observability and control across data, compliance, identity, access, and security surfaces rather than isolated application logs.
For a human worker, the enterprise usually has a mature joiner, mover, and leaver structure: employee record, manager, role, cost center, access review, termination workflow, and audit expectations.
For non-human actors, those joins are less natural.
The owner may be a project team. The credential may be an application registration. The action may be a vendor-run agent. The business effect may land in a human-centered workforce system.
The risk is not only over-permissioning. It is audit fragmentation. The enterprise can know pieces of the story without owning the whole record.
Offboarding follows from the missing record.
This is where Part II links back to Brief 005.
Agent authority can persist, expand, or be repurposed after the original task, owner, workflow, or business context changes. That is the lifecycle failure.
But the lifecycle failure becomes harder to solve when the agent is absent from the workforce record.
If the organization cannot identify the agent as an accountable non-human actor, it cannot reliably determine who owns revocation, what authority should expire, what actions remain auditable, or which system should carry the surviving record.
Offboarding for agents is therefore not the headline of Part II. It is the consequence.
A workforce record is what makes agent offboarding governable.
Watch for these signals that the market is moving from agent identity toward agent workforce records.
- HRIS, LMS, recruiting, case-management, or workforce-automation records show a business action but do not preserve the non-human actor that produced or influenced it.
- Agent inventories live in IAM or vendor tools but do not join cleanly to workforce ownership, manager context, business purpose, or work output.
- Access reviews can identify a credential but cannot identify the agent’s current owner, approved purpose, business context, or authority-expiry trigger.
- HR, learning, workflow, GRC, and security logs each hold partial evidence of agent action without a single accountable lifecycle record.
- Procurement and risk reviews ask how an agent is provisioned, but not where it is recorded as a workforce actor or how its action trail survives ownership change.
- Vendors add owner fields, agent registries, audit trails, or non-human worker concepts to workforce, identity, workflow, or governance systems.
- Incident response teams can suspend a credential but cannot reconstruct which workforce process the agent acted in or under whose authority.
For CHROs and HR technology owners.
Agent deployment should be treated as a workforce-record question before it becomes an operating norm. If an agent touches employee data, recruiting steps, HR cases, learning assignments, approvals, or performance-adjacent workflows, the enterprise needs an actor record that survives beyond the transaction.
For CISOs and identity leaders.
The test is not only whether the agent has least privilege. It is whether the identity record joins to workforce accountability: owner, purpose, scope, business context, action history, review trigger, and revocation evidence.
For CLOs and learning-system owners.
The near-term risk is not a broad claim that learning records are already polluted. That remains under-evidenced. The defensible control question is narrower: when an agent participates in learning or workforce workflows, can the system separate the learner, manager, application, vendor, and non-human actor in the record?
For GRC, legal, and audit teams.
The evidence burden is cross-system. The transaction log is not enough. Reviews should ask whether the enterprise can reconstruct the accountable non-human actor across HRIS, LMS, IAM, workflow, GRC, SIEM, and vendor systems.
For vendors and procurement teams.
The minimum ask is no longer “does this agent integrate?” It is “where does the agent appear in the workforce record, who owns it, what authority does it carry, how does it expire, and what evidence proves what it did?”
The strongest objection is that this is still an identity-governance problem.
Mature IAM, PAM, service-account governance, workload identity, access reviews, policy engines, and security telemetry already handle non-human actors. A separate workforce-record frame could add process language without improving control.
That objection is partly right.
Part II should not imply that HRIS is always the correct source of truth, that every agent needs employee-style treatment, or that identity/security controls are secondary. For many technical agents, the better control surface may remain workload identity, application governance, SIEM, GRC, or a vendor control plane.
The response is that workforce action creates a different proof burden.
When an agent influences work involving employees, learning, HR, approvals, cases, skills, or knowledge workflows, the enterprise needs a record that can answer workforce-accountability questions, not only authentication questions.
There is also an evidence limit.
The current support can carry a qualitative argument about workforce recordability and cross-system traceability. It cannot carry claims that all HR systems fail, that agent participation is already distorting learning measurement at scale, that executives must redesign all decision rights, or that every machine-identity failure is an org-chart failure.
The narrower claim is stronger: when agents perform workforce work, the enterprise needs a durable record of the non-human actor. Without that record, Brief 005’s authority problem becomes harder to govern.