TL;DR: C1.ai says identity deployments can now start from a goal, read the live tenant, and assemble governance tasks, AI access setup, and approvals through agentic onboarding rather than a manual setup project. The deeper issue is that onboarding is becoming execution, so approval gates, evidence trails, and control design move earlier in the lifecycle.
At a glance
What this is: C1.ai describes an agentic onboarding flow that builds identity governance setup from a stated goal, using live tenant data to assemble tasks, approvals, and AI access controls.
Why it matters: This matters because IAM and IGA teams may need to govern onboarding as an automated execution path, not just a configuration project, especially as AI access and MCP tools are added on top of people governance.
👉 Read C1.ai's blog post on agentic onboarding for identity governance
Context
Identity onboarding is traditionally a configuration-heavy exercise: teams connect directories, map attributes, define workflows, and then wait for value to appear. The article argues for a different model in which the onboarding flow begins with an outcome, reads the live environment, and assembles the work needed to reach it.
That shift matters for identity governance because the setup process itself starts to carry control decisions, evidence collection, and approval logic. For IAM, IGA, and emerging AI access programmes, the question is no longer only whether controls exist, but when the system is allowed to help create them and under what guardrails.
The article also links people governance to AI access governance, placing AI access on top of a foundation built for human identities. That sequencing is important because it treats agent access as dependent on existing governance maturity rather than as a separate, isolated rollout.
Key questions
Q: What breaks when onboarding itself can execute identity setup tasks?
A: What breaks is the assumption that onboarding is only a pre-production configuration step. When the onboarding flow can read live tenant data, assemble tasks, and trigger changes, it becomes part of the control plane. Teams need to govern who authorises that execution, what data it can inspect, and which steps remain human-approved.
Q: Why do agentic onboarding flows increase governance risk if they feel easier to use?
A: Because ease of use can hide the moment when the system stops collecting inputs and starts making operational decisions. If approval gates, evidence capture, and role mappings are created during onboarding, weak defaults get embedded earlier. That is not just convenience, it is governance scope moving left into implementation.
Q: How should teams govern MCP tools discovered during AI access onboarding?
A: Treat every discovered MCP tool as an entitlement that needs classification, approval, and ownership before use. If a connected app exposes tools to agents, the governance model should say who can approve them, what data they can reach, and whether new tools default to restricted access until reviewed.
Q: How does agentic onboarding change the relationship between human IAM and AI access governance?
A: Human identity governance becomes the foundation rather than a separate track. If directory data, manager relationships, and employment status are not clean, the AI access layer inherits those defects. Teams should therefore sequence people governance first, then allow agent access to build on that baseline.
Technical breakdown
Goal-driven onboarding and tenant discovery
The article describes onboarding as a runtime process that begins with a desired outcome, such as audit remediation, and then inspects the tenant to determine the next steps. Instead of a static wizard, the system detects the identity provider, reads reporting relationships, employment status, app scope, and other existing state, then turns that into an ordered plan. Technically, this is discovery plus orchestration: the platform uses live identity data to scope work, prefill configuration, and route tasks in sequence. That reduces manual setup, but it also means the onboarding flow is now making decisions based on environment state rather than simply collecting inputs.
Practical implication: teams should treat onboarding logic as governed automation and review what tenant data it is permitted to read before trusting its recommendations.
Agent-assisted execution with approval gates
The article distinguishes between agent-assisted tasks and tasks that still require user validation. In this model, the agent assembles a campaign or executes setup work, but sensitive changes remain approval-gated. That is an important control pattern: the system can reduce coordination overhead without removing the human decision point for consequential actions. The practical question is not whether an agent can do the work, but which steps can safely be delegated, which steps must remain advisory, and how the approval trail is preserved. This is closer to supervised orchestration than full autonomy, even though the experience feels agentic.
Practical implication: define which onboarding tasks can be executed automatically, which must be proposed, and which require explicit human approval before environment changes occur.
Agentic access management for MCP tools
The article extends onboarding into AI access management by discovering MCP servers exposed by connected apps and applying default governance to them. MCP, the Model Context Protocol, connects AI agents to tools and data sources, so governance now has to cover tool discovery, tool classification, and tool approval before use. The article says discovered tools default to sensitive classification and require approval before use. That means access control is not only about identities and apps anymore. It also has to account for the tools an agent can reach, which is a separate governance surface from classic directory-centric IAM.
Practical implication: inventory MCP-connected tools as part of onboarding and enforce approval-before-use controls for newly discovered AI access paths.
NHI Mgmt Group analysis
Goal-led onboarding is collapsing the distance between configuration and execution: the article shows that identity setup no longer has to be a human-run project that ends in a finished configuration. When the onboarding flow reads tenant state, proposes tasks, and can carry some of them out, the setup layer becomes part of the control plane. Practitioners should stop treating onboarding as pre-governance administration and start treating it as governed execution.
Identity governance now starts before the first review cycle: the article makes clear that access review, manager routing, and evidence capture are being introduced during onboarding rather than after deployment. That changes how control maturity should be judged, because the first meaningful governance decision may happen while the system is still assembling itself. The implication is that evidence quality and approval logic now matter inside implementation, not just after go-live.
People governance is becoming the prerequisite for agent governance: the article's sequencing is explicit, with human identity governance first and AI access layered on top. That ordering is sound because AI access inherits whatever directory truth, role mapping, and approval discipline the human layer already has. If the base identity model is weak, agent governance becomes a projection of that weakness, not a compensating layer.
MCP creates a new governance surface for identity teams: once connected applications expose MCP servers, the question is no longer only who can log in, but which tools an agent can discover and use. The article's sensitive-by-default stance is the right category boundary to notice, because tool exposure is now an identity issue, not just an application integration detail. Teams should treat discovered tools as governed entitlements, not implementation noise.
Agentic onboarding is a named concept worth tracking: onboarding that assembles, validates, and sometimes executes its own setup work is no longer a future-state idea, it is an operational pattern. That pattern compresses deployment time, but it also compresses the time available to inspect assumptions, evidence, and approval boundaries. The practical conclusion is that onboarding design has become an IAM governance problem, not merely a product usability question.
What this signals
Agentic onboarding shifts control design left: if setup can assemble governance tasks from live tenant data, then identity programmes need to decide where configuration ends and execution begins. The operational boundary is no longer obvious, so approval logic and evidence capture have to be designed into the onboarding path itself.
A useful way to think about this is that onboarding is becoming a governance workbench, not a provisioning checklist. That means practitioners should look for where their current process still assumes a human will translate discovery into action later, because that assumption is shrinking fast.
For practitioners
- Define onboarding as a governed workflow Map which setup steps are informational, which are decision points, and which can be executed automatically during identity deployment.
- Review approval gates before agentic execution Require explicit approval for any onboarding step that changes access, creates campaigns, or applies governance actions to live accounts.
- Treat tenant discovery as sensitive data access Limit which identity attributes, reporting lines, and employment-status fields the onboarding flow can read before it builds a plan.
- Inventory MCP tools as entitlements Track every discovered MCP server and its exposed tools as part of the access model, with sensitive-by-default handling for new tools.
- Separate human and agent governance milestones Sequence people governance before AI access so the agent layer inherits a stable directory, clean role mapping, and auditable approvals.
Key takeaways
- The article shows identity onboarding moving from manual setup toward goal-driven execution that can inspect tenant state and assemble governance work automatically.
- The practical shift is not just speed. Approval gates, evidence trails, and tool classification now sit closer to the start of the identity lifecycle.
- For practitioners, the key question is whether onboarding is still a configuration process or has become part of the control plane that needs its own governance model.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST CSF 2.0 and NIST Zero Trust (SP 800-207) set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | Agentic onboarding delegates setup actions under controlled permissions. |
| Recommendation — Bound agentic onboarding to least-privilege execution and approval gates for any identity-changing action. | ||
| OWASP Non-Human Identity Top 10 | NHI-04 — Insecure Authentication | The article centers on AI access and discovered tools needing approval before use. |
| NHI-05 — Overprivileged NHI | Onboarding must limit what agentic workers and AI connections can reach. | |
| Recommendation — Apply NHI-04 controls to discovered AI access paths and require approval before tool use. Review agent and AI access scope for overprivilege before enabling onboarding-driven execution. | ||
| NIST CSF 2.0 | PR.AA-05 — Access Permissions, Entitlements and Authorizations | The post is about setting and governing permissions during onboarding. |
| Recommendation — Use PR.AA-05 to govern entitlement assignment as part of the onboarding workflow. | ||
| NIST Zero Trust (SP 800-207) | Access control principles — Access control principles | The workflow relies on continuous authorization decisions across connected systems. |
| Recommendation — Apply zero trust principles to verify each onboarding action before it can change access. | ||
Key terms
- Agentic onboarding: An onboarding process in which software assembles and executes setup tasks from a goal rather than requiring an administrator to drive each step manually. In identity governance, this matters because the setup path itself becomes part of the control environment and must remain auditable.
- MCP Server: An MCP server is a tool endpoint that connects an AI agent to external systems and data sources through Model Context Protocol. Because it extends what the agent can reach, it becomes part of the identity and access surface and must be reviewed like any other privileged connector.
- Approval Gate: An approval gate is the human or policy checkpoint that must be crossed before an AI system can take a state-changing action. For identity governance, the gate is the point where responsibility stays with the organisation, and it should be enforced separately from the system’s ability to observe or recommend.
- Tenant Discovery: Tenant discovery is the step where a system determines which organisation a user intends to access after identity is proven. It may use a subdomain, invite link, domain mapping, or org picker. In SaaS, discovery is part of authorisation because the same user can belong to multiple tenants with different policies.
What's in the full article
C1.ai's full blog post covers the operational detail this post intentionally leaves for the source:
- The step-by-step onboarding sequence from goal selection through directory connection and tenant discovery
- The live examples of how audit remediation, access reviews, and manager coverage checks are assembled during setup
- The mechanics of agent-assisted task execution and where user approval remains mandatory
- The AI access management flow for MCP servers, including how discovered tools are surfaced and classified
Deepen your knowledge
NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.
Published by the NHIMG editorial team on August 22, 2026.
Updated on October 8, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org