Agentic AI, independent identities, and the next growth wave for software vendors

Agentic AI is changing one of the most basic assumptions in enterprise software: that the user of a software application is a human being sitting in front of a screen.

AI agents are different. They can interpret goals, access tools, interact with APIs, move across workflows, retrieve data, trigger actions, and in many cases operate with a degree of autonomy. This is exactly what makes them powerful. It is also what makes them risky.

When an AI agent can act inside enterprise systems, the question is no longer only “What can this user do?” It becomes: Which agent is acting, on whose behalf, with which permissions, for which purpose, and under which commercial entitlement?

That question sits at the intersection of cybersecurity, identity management, software licensing, and digital business strategy. For software vendors, it also points to a promising future: As enterprises adopt agentic AI more broadly, they will need stronger mechanisms to secure, govern, meter, and monetize access to software capabilities.

 

AI agents are becoming software actors

The rise of agentic AI is not just another step in automation. Traditional automation usually follows predefined rules. AI agents can operate more dynamically: They can decide which tool to use, adapt to context, and perform multi-step tasks across systems.

This changes their role in the software ecosystem. An AI agent is no longer merely a background script or a passive assistant. It can become an active software actor.

That shift is already visible in the way public institutions and major technology providers are approaching the topic. NIST has launched work around AI agent standards, with explicit attention to identification, authorization, auditing, non-repudiation, and controls against prompt injection. Microsoft Entra Agent ID now defines agent identities as a way to distinguish AI agent operations from human, customer, or workload identities and govern their access accordingly.

The message is clear: If AI agents are going to act in enterprise environments, they need to become visible, governable, and accountable.

 

Why shared human credentials are the wrong foundation

The simplest way to let an AI agent use enterprise software is to let it act under a human user’s account. It is also one of the most problematic.

Shared or delegated access without clear agent identity creates several risks. It becomes harder to know whether a specific action was performed by the human user, the AI agent, another automation layer, or a compromised process. Permissions can become too broad. Audit trails can lose meaning. Incident response becomes more complex because the organization cannot easily distinguish legitimate human behavior from autonomous agent behavior.

Security agencies are now warning organizations to treat agentic AI as a serious cybersecurity topic. Recent guidance supported by agencies including the NSA recommends incremental deployment, continuous threat-model assessment, strong governance, explicit accountability, monitoring, logging, and human oversight.

In practice, this means that AI agents need their own managed identities, not just borrowed credentials. They need to be authenticated, authorized, monitored, and revoked when no longer needed. They need to operate under policies that are more precise than a conventional human user profile.

 

From least privilege to least agency

Least privilege has long been a cornerstone of cybersecurity. In an agentic AI environment, it needs to evolve.

For human users, least privilege usually means granting only the access rights required for a person’s role. For AI agents, the question is broader. It is not just what the agent can access, but what the agent is allowed to do.

Can it read data but not modify it? Can it create a quote but not approve a discount? Can it run a simulation but not export the resulting model? Can it activate a software feature but not extend a license? Can it act only during a specific workflow, only for a specific customer, only after human approval, or only within a defined budget?

This is where independent agent identities become a practical security control. Once an agent is identifiable, administrators can assign granular entitlements to it. They can define what the agent may access, which features it may use, which APIs it may call, how often it may act, and when its authority expires.

In other words, identity is the foundation; entitlement is the control layer.

 

Security policies need runtime enforcement

Policies alone are not enough. Agentic AI introduces dynamic behavior, and dynamic behavior requires runtime enforcement.

An enterprise can define the perfect policy on paper, but the critical question is whether that policy is enforced when the agent actually calls a tool, consumes a feature, modifies a dataset, launches an operation, or interacts with another system.

This is why agentic AI governance cannot stop at identity management. It also needs enforceable rights, usage limits, logging, and revocation mechanisms. Each action should be evaluated against the agent’s current authorization context.

For software vendors, this is a familiar challenge. Protecting and licensing software already means controlling who or what can access which capabilities, under which conditions, and for how long. Agentic AI simply expands the number and variety of actors that need to be controlled.

 

The licensing question: who is the real user?

Once AI agents have independent identities, the licensing question follows naturally.

If a human user interacts with a software application through an AI agent, is the software being used by the human, the agent, or both? If one employee uses five agents to perform specialized tasks across different tools, is that still one user? If a team deploys hundreds of ephemeral agents to perform data analysis, simulation, configuration, testing, or support workflows, should access be licensed per person, per agent, per task, per feature, per transaction, or per outcome?

There is no universal answer. Microsoft’s current Agent ID documentation, for example, states that agents acting on behalf of a licensed user are covered under that user’s license and do not require their own separate license, while some security features for agents require additional licensing. That is an important reminder: the market will not move toward one simplistic licensing model.

But the broader trend is unmistakable. Deloitte has explicitly connected agentic AI with changes in identity and access management, software licensing, entitlement, and provisioning. As agents become more capable, more autonomous, and more numerous, software vendors and enterprise customers will need more sophisticated commercial models.

The future is unlikely to be “one agent, one license” in every case. It is more likely to include a mix of:

• Delegated-user licensing, where the agent acts within the rights of a licensed human user
• Named-agent licensing, where persistent agents are treated as distinct licensed actors
• Pooled agent licenses for teams, departments, or business units
• Feature-based licensing, where agents can access only specific capabilities
• Usage-based licensing for high-volume or transactional agent activity
• Time-limited entitlements for temporary or project-specific agents
• Policy-bound licenses that combine security conditions with commercial rights

This is not just a compliance burden. It is a new opportunity to align software value with actual software consumption.

 

Why this is good news for software vendors

Agentic AI will increase the demand for software. It will also increase the demand for controlled software access.

As enterprises automate more workflows with AI agents, more software capabilities will be triggered more frequently, in more combinations, and by more actors. Software that was once used occasionally by a limited group of expert users may become part of continuous AI-assisted processes. Specialized features that were previously hard to expose to a broad audience can become accessible through controlled agent-driven workflows.

That creates new growth opportunities for software vendors.

First, vendors can monetize software more precisely. Instead of relying only on traditional seat-based licensing, they can offer commercial models that reflect how customers actually use software in agentic workflows.

Second, vendors can reduce hidden or uncontrolled use. Independent identities and enforceable entitlements make it easier to distinguish legitimate access from overuse, misuse, or unauthorized automation.

Third, vendors can offer AI-ready licensing as a competitive advantage. Enterprises adopting agentic AI will need suppliers that can support granular access, secure delegation, auditability, and scalable entitlement models.

Fourth, vendors can turn security into a business enabler. Stronger identity and entitlement controls do not simply reduce risk. They make it possible to open software capabilities to AI-driven processes with confidence.

 

The enterprise perspective: more control, but also more cost

For enterprise customers, the implications are more mixed.

Independent agent identities improve security, governance, and accountability. They make it possible to limit what agents can do and to investigate what happened when something goes wrong. They help organizations avoid the dangerous ambiguity of AI agents operating through broad human accounts or unmanaged service credentials.

At the same time, agentic AI may increase software costs. If agents become frequent consumers of licensed capabilities, enterprises may need additional entitlements, higher usage tiers, new licensing models, or expanded governance features. Multiple agents per employee could create licensing complexity that traditional procurement and SAM processes were not designed to handle.

This is not necessarily a negative development. It is a sign that software is being used more deeply and more productively. The important point is transparency. Enterprises need to understand what they are licensing, how agents consume software, and which security controls are included.

The vendors that make this clear will have an advantage.

 

CodeMeter’s natural role in the agentic AI landscape

The agentic AI discussion reinforces a principle that has always been central to modern software protection and licensing: access to software capabilities should be controlled, enforceable, measurable, and adaptable.

In an AI-driven enterprise, that control must extend beyond the traditional human user. It must support a wider landscape of actors: people, devices, machines, services, cloud workloads, and increasingly AI agents.

This is where technologies like CodeMeter fit naturally. CodeMeter enables software vendors to protect intellectual property, define granular license rights, enforce access to features, support flexible business models, and manage entitlements across different deployment environments.

As AI agents become part of enterprise workflows, licensing can evolve from a static commercial mechanism into an active governance layer. It can help define not only whether software is paid for, but also whether a specific actor is authorized to use a specific capability in a specific context.

That is the real opportunity. Licensing becomes part of secure digital value creation.

 

From security necessity to business growth

Agentic AI will not eliminate the need for licensing. It will amplify it.

The more software is embedded into AI-driven workflows, the more important it becomes to know who or what is using it. Independent agent identities will help enterprises secure autonomous behavior. Granular entitlements will help administrators define what agents are allowed to do. Runtime enforcement will help ensure that policies are respected. And flexible licensing will help software vendors monetize the expanded use of their capabilities.

This is a positive outlook for the software industry. Agentic AI will create new risks, but it will also create new demand. Vendors that can combine security, entitlement management, and licensing flexibility will be well positioned for the next phase of digital business.

The future of software licensing will not be limited to counting human users. It will be about governing every authorized actor that creates value with software.

And increasingly, some of those actors will be AI agents.