Agentic AI shifts enterprise software logic to runtime, not rules

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Saviynt's blog post: Agentic AI is changing enterprise software from predefined workflows to runtime decision-making, exposing the limits of identity governance and segregation of duties controls built for periodic review.

Why it matters: IAM, IGA, and PAM teams need to reassess how they enforce access, approval, and accountability when the decision point moves into execution rather than sitting upstream in a rules engine.

Context

Agentic AI is pushing enterprise software away from fixed business rules and toward runtime decisions. In this model, the system no longer waits for humans to encode every branch in advance; it evaluates context as work happens and acts within guardrails. That changes the identity governance problem because many controls still assume access and conflict states can be reviewed after provisioning.

The governance gap is especially visible in Identity Governance and Administration and segregation of duties. Those processes were designed for periodic synchronisation, exception handling, and after-the-fact certification, not for a system that can decide and execute in the same moment. The article’s central claim is that the software layer is no longer just a workflow boundary; it is becoming the place where governance must occur.

Key questions

Q: How should security teams govern access when business logic moves to runtime?

A: Security teams should move from after-the-fact review to point-of-decision control. When agents or runtime systems can evaluate context and act immediately, the governance model has to constrain what they are permitted to decide, not just what they are permitted to request or review later.

Q: Why do agentic AI systems break segregation of duties models?

A: They break SoD models because the conflict can be created and consumed inside one runtime sequence. A control that checks access after provisioning may never see the prohibited combination if the system finishes the task before review. SoD for agentic systems has to govern sequences and decisions, not just static permissions.

Q: What breaks when identity governance still depends on periodic certification?

A: Periodic certification breaks when identities change faster than the review cycle can capture. Access drift, standing privilege and delegated permissions can remain in place long after the business context has changed. In mixed human, non-human and AI estates, that means governance can look complete on paper while actual exposure keeps growing between review windows.

Q: What is the difference between agentic AI governance and traditional automation governance?

A: Traditional automation governance assumes a fixed script with predictable inputs and outputs. Agentic AI governance must control a system that can plan, adapt, and choose different execution paths. That means identity scoping, approval rules, and audit trails matter more, because the same agent may behave differently across similar tasks.

Technical breakdown

Why runtime decision-making breaks rules-engine governance

Traditional enterprise software depends on predefined logic: if a condition is true, a rule fires, a workflow advances, and the system moves on. Agentic AI changes that by making decisions in context at execution time rather than relying on a fixed branch map. That matters because rules engines, approval chains, and exception queues are all built on the assumption that decisions can be modelled before the work starts. Once the decision is made during the task itself, the control point shifts from configuration to execution, and the old governance boundary becomes too late to be authoritative.

Practical implication: Map which controls still assume pre-declared decision trees and identify where runtime authority now sits.

How identity governance and segregation of duties lose timing control

Identity Governance and Administration works well when entitlement states change slowly enough to be discovered, reconciled, and certified. Segregation of Duties policies also rely on a visible window in which conflicting access can be detected before risk becomes irreversible. The article’s example of a source-system change arriving before the IGA platform updates shows the timing problem clearly. In agentic workflows, that delay becomes structural rather than incidental, because the system itself can continuously interpret context and proceed before periodic governance cycles catch up.

Practical implication: Shift governance checks closer to issuance and decision time instead of depending on delayed certification cycles.

Why outcomes matter more than tools in agentic software

The article argues that software is moving from tool-centric design to outcome-centric execution. In that model, humans define intent, constraints, and guardrails, while the system decides how to complete the work. That is not the same as simple automation, because the sequence is not fixed in advance and the execution path can adapt in real time. For identity teams, the implication is that accountability has to attach to the permitted outcome and the decision context, not only to the application or workflow that used to enforce the step-by-step process.

Practical implication: Review governance models for outcome-based execution and determine where approval, exception handling, and logging need to move.

NHI Mgmt Group analysis

Agentic AI turns identity governance from a periodic control into a runtime control problem. The article is not describing a marginal workflow improvement; it is describing a change in where the decision is made. Identity governance was built for systems that needed time to synchronise, review, and certify. Once the system itself can evaluate context and act during execution, that governance posture becomes too slow to be decisive. Practitioners should treat runtime authority as the new control boundary.

Segregation of duties breaks when the conflict appears after the action path is already underway. The article’s IGA example shows the classic failure mode: a policy exists, the tool knows the rule, but the enforcement window lags the transaction. Agentic software collapses that window further because the logic moves into the work itself. That means the discipline is no longer about proving the rule exists on paper; it is about proving the decision path is constrained at the moment of action.

Business applications are losing their role as the place where human intent is translated into fixed instructions. That translation layer was the assumption underlying much of enterprise governance, including role design, workflow approvals, and exception handling. When agents own more of the decision process, the governance question changes from who approved the workflow to what the system is allowed to decide on its own. The implication is a redesign of identity control boundaries, not a tuning exercise.

Runtime governance becomes the named concept that matters here: controls must move to the point where the decision is made. The article shows that the old sequence of request, sync, review, and certify is no longer adequate when execution and evaluation converge. This is a structural shift across AI governance, IGA, and enterprise application design. Practitioners should assume that any control depending on delayed visibility will underperform in agentic environments.

The control model is shifting from enforcing rules about users to constraining decisions made by systems. That is the deeper implication for identity security. Humans remain accountable, but the system is now the actor that reaches operational outcomes, and that changes how access, exceptions, and approvals must be governed. The practical conclusion is that identity programmes need to measure where decision authority lives, not just who can log in.

From our research library:

• Gartner predicts that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024, and that 15% of day-to-day work decisions will be made autonomously.
• Read next: Agentic AI Identity Maturity Model

What this signals

Runtime governance gap: Identity programmes built on delayed certification are being asked to govern systems that decide in the moment. That means the control plane has to move closer to execution, especially where access and segregation of duties are still validated on a schedule rather than at the point of action.

Only 13% of organisations feel extremely prepared for the reality of agentic AI despite the majority racing toward autonomous adoption, according to the 2026 Infrastructure Identity Survey. That gap suggests governance teams will be forced to redesign controls while the operating model is already changing.

For practitioners

• Inventory runtime decision points Map where enterprise applications now evaluate context and make decisions during execution rather than through prebuilt rules. Prioritise workflows where identity, access, or segregation of duties controls still assume a delayed review window.
• Rebuild segregation of duties around action time Identify conflict checks that currently fire after provisioning or after sync, then move them toward issuance and execution so the system cannot complete a risky action before governance sees it.
• Separate approval from autonomy Define which outcomes an agentic system may complete without a human gate, which require escalation, and which must remain outside automated execution entirely.
• Measure governance lag as a security signal Track how long it takes for identity, entitlement, and policy state to reflect a change in source systems, then use that lag to identify where runtime controls are being bypassed by design.

Key takeaways

• Agentic AI changes the governance problem by moving business logic into runtime, where periodic identity controls arrive too late to be authoritative.
• The most exposed programmes are those that still depend on synchronisation, certification, and approval chains to validate access after the work has already begun.
• Identity teams need to measure where decision authority now sits and redesign controls around execution time, not just around policy design.

Key terms

• Agentic AI: Autonomous AI systems capable of planning, deciding, and taking actions, including calling APIs, writing code, and orchestrating other agents, with minimal human oversight. Agentic AI introduces new NHI risks as agents must authenticate to external services.
• Segregation of Duties: Segregation of Duties is a control principle that prevents one person or role from combining incompatible permissions that could create fraud, error, or undetected change. In ERP environments, it must account for roles, transactions, approvals, and compensating controls across business processes.
• Runtime Governance: Runtime governance is the set of controls that verify what a system or agent is actually doing after deployment. It combines monitoring, authorization checks, and access validation so teams can detect drift, misuse, or excessive privilege in motion rather than assuming build-time policy still holds.
• Outcome-Based Workflow: An outcome-based workflow is designed around the result the business wants, not around a fixed application path. In identity terms, this shifts governance away from prebuilt screens and approval chains toward policies, guardrails, and traceable decisions that can adapt as the task unfolds.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity security 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.