Agentic AI Is Here. Now We Need a Control Plane for the Agents.

Written by David (DJ) Morimanno, Field CTO, North America at Xalient

4 December 2025

The original article captured something many organizations have not fully acknowledged yet. Agentic AI is not a trend or another automation wave. It is a structural shift in how work is executed, how systems interact, and how decisions get made. It reshapes the architecture of the enterprise, and more importantly, it reshapes the identity landscape itself.

But as compelling as that vision is, there is a parallel truth that must be stated just as strongly. The industry is racing ahead on capability, and almost no one is talking about control. The real question is no longer about whether to adopt Agentic AI. The real question is, who governs the agents, who defines what they are allowed to do, and how do we maintain authority when they begin acting on our behalf?

Agentic AI is not a traditional application. It is not a static workload. These systems operate with intent. They perceive, plan, and execute. They cross boundaries. They chain actions together. They move at a speed and scale that most organizations are not prepared to supervise or contain.

Governance cannot lag behind this shift. If it does, the risks grow exponentially.

Agentic AI Changes the Risk Landscape

Every AI agent is, in effect, a new type of identity. It reads data, touches systems, and triggers workflows. That means it carries privileges. It carries entitlements. And if those permissions are not governed with clarity and precision, its blast radius will expand quietly in the background.

We are already battling the accumulated weight of decades of unmanaged access. Over-privileged service accounts, inherited permissions, static keys, broad-role entitlements, and blind spots in cloud exposure. If Agentic AI is allowed to grow without discipline, we will recreate all of those problems at a scale and velocity we have never seen before.

The issue is not with the models. The issue is with the absence of a unified approach to defining, constraining, supervising, and revoking agent permissions. Organizations lack a shared model for:

  • Who authorizes an agent
  • What the agent is allowed to access
  • How its permissions are defined
  • How its actions are monitored
  • How access is revoked if something goes wrong

This is no longer about automation. It is about authority and control.

Governance Must Evolve with the Technology

As organizations expand their use of autonomous agents, governance becomes a strategic requirement. Regulators are already driving expectations around transparency, accountability, explainability, and lifecycle management. Business leaders, meanwhile, are pushing for speed, automation, and operational efficiency.

To bring these forces together, governance must be rebuilt around identity, access, and privilege. AI agents do not simply run tasks. They interpret goals, make decisions, and act across systems. That makes them powerful, but also uniquely high-risk.

The solution is to formalize how these agents are governed and treated as identities within existing security and identity programs.

Treat Agents as First-Class Identities

Key Agentic AI Categories

  1. Personal Assistant Agents: Microsoft Copilot exemplifies agentic help, integrated across Microsoft 365 for content generation, meeting summaries, document drafting, and more. Google’s Astra is another, offering multimodal capabilities akin to a Star Trek computer.

  2. AI Agents for Writing Code: GitHub Copilot, Tabnine, Codeium (Qodo), and Replit are getting so good now that applications can be written through a simple prompt in natural language. Satya Nadella, Microsoft CEO, notes that 20–30% of all code is already written by AI (10:30 into this video), a figure he believes will rise.

  3. AI Agents for Website Design: Tools like Tars Web Builder, 10Web, and Wix ADI create professional websites from simple prompts, handling Search Engine Optimization (SEO), imagery, and content. It is easy to predict that Agentic Engine Optimization (AEO) will become a new buzz phrase.

  4. Workflow Automation Agents: Zapier and Microsoft Power Automate connect apps, automate workflows, and provide enterprise-level process automation.

  5. Research & Analysis Agents: Perplexity and OpenAI’s Operator deliver concise, cited answers for complex queries, combining retrieval-augmented generation and advanced context management.

  6. Customer Service Agents: Ada and Sierra use natural language understanding and sentiment analysis to provide efficient, always-on support, handling everything from FAQs to escalations.

  7. Sales and Marketing Agents: Artisan AI automates lead qualification, pipeline management, and campaign optimization, leveraging behavioral analytics and CRM integration.

  8. Data Analysis Agents: Tableau Pulse and Microsoft Fabric/Power BI Copilot enable conversational analytics, dynamic dashboards, and predictive insights via natural language queries. Many enterprises can do this now with Copilot—just ask it to visualize a file for you, and in less than a minute you have an interactive web report.

  9. Creative & Content Agents: Copilot has a podcast generator that builds a 5–6-minute dialogue between a ‘virtual reporter’ and ‘virtual subject expert’ on any topic you prompt it with. Runway Gen-3 Alpha and Sora generate video and imagery from prompts. Check out the “Will Smith Eating Spaghetti Test,” and you can see how far this video creation has come in two short years.

  10. Specialized Domain Agents: Legal agents for advice and content support; DevOps agents automate deployment; healthcare agents analyze scans and manage records; finance agents detect fraud and optimize investments; education agents create adaptive learning paths and grade assessments. Some agents even manage teams of other agents.

    This is the foundation of what becomes the AI Access and Control Plane.

The AI Access and Control Plane

As Agentic AI becomes embedded across business processes, the enterprise needs a unified control model that:

  • Governs every actor, whether human, machine, or AI
  • Defines access based on intent, identity, and policy
  • Monitors behavior in real time
  • Enforces Zero Trust across identity, data, and network boundaries
  • Provides accountability and auditability
  • Connects AI governance with IGA, PAM, CIEM, ITDR, and secure networking

This is the missing layer in most AI strategies today. The capabilities are advancing rapidly, but the control model is not advancing at the same pace. The risk surface is expanding faster than the guardrails designed to contain it.

A Practical First Step

Organizations do not need an enormous program to begin. They need clarity, visibility, and structure.

A practical starting point includes:

  • Identifying all agents already in use
  • Understanding their permissions and mapping their risk
  • Establishing a governance baseline
  • Designing a controlled permission model for one or two key agents
  • Folding this model into existing identity and Zero Trust programs
  • Turning early lessons into reusable policy and architecture

This approach accelerates innovation while maintaining organizational control.

 

Where Xalient Supports the Journey

This challenge aligns naturally with the work Xalient has been delivering for years. Our expertise in identity, secure networking, and Zero Trust positions us to guide organizations through this next stage of transformation.

We help organizations:

  • Build AI governance frameworks
  • Model AI agents as identities
  • Design permission structures for autonomous systems
  • Align agent workflows to Zero Trust principles
  • Monitor agent behavior and detect drift
  • Integrate AI activity into ITDR and threat detection programs
  • Provide managed governance for organizations looking to scale safely

Agentic AI will reshape how enterprises operate. That is no longer in question. The question is whether organizations deploy it with the right guardrails and the right oversight.

Now is the moment to build the control plane that ensures AI remains governed, accountable, and aligned to business intent.

 

Picture of David (DJ) Morimanno, Field CTO, North America at Xalient

David (DJ) Morimanno, Field CTO, North America at Xalient

DJ helps clients develop IAM strategies that work in complex organisations. He’s an active practitioner and strategist, with nearly 20 years of hands-on experience in implementing market-leading IAM technologies across IGA, PAM, and Access Management. He specialises in building IAM Programs, administering IAM tools, and developing long-term strategies to support organisational objectives and business enablement. 

DJ has a passion for cybersecurity.  He is a trusted advisor for Fortune 500 clients and has helped industry executives successfully execute large-scale IAM programs through deployment.  He has extensive experience in financial services, energy, education, manufacturing, and healthcare industries. 

Search

Categories

HIGHLIGHTS

Follow us

Linkedin X-twitter Facebook-f Youtube

Speak to an Expert

Explore the power of Xalient Solutions

Get in Touch
close menu icon

Subscribe to our Newsletter!