While AI adoption soars globally, executives and researchers from leading AI developers have warned that the commercial race to achieve artificial general intelligence (AGI) is moving faster than the world’s ability to build safety boundaries. To balance staying abreast in the AI race with mitigating new risks posed by AI, organisations must formalise AI governance.
This is according to Thabiso Hlatshwayo, Director – Solution Consulting and Growing Markets at OpenText, who says taking control of AI requires a platform approach with a central control plane.
A new phase of AI
Hlatshwayo says: “We are entering a new phase of enterprise AI. The first wave of generative AI was largely about helping people find, understand and create information. Agentic AI takes this further. Agents can increasingly take action, interact with enterprise systems and execute repeatable business tasks on behalf of people. That shift changes the conversation. The question is no longer simply, ‘What can AI do?’ It is increasingly, ‘How do we control what AI is allowed to do?’"
He notes that agentic AI can impact operations and functions across enterprises.
In enterprise content management (ECM), an agent could classify a document, update metadata, move content, apply permissions or initiate a records-management action.
“These are no longer theoretical scenarios,” he says. “For example, AI supports agentic actions such as creating workspaces, updating metadata, moving documents, changing permissions, classifying records and applying legal holds.”
In IT service management (ITSM), agents can support repeatable activities around incidents, changes and knowledge. AI material highlights agent workflows for areas such as change post-implementation reviews, incident-title generation, knowledge creation and change-risk analysis.
“In data and information governance, the opportunity becomes equally significant,” he says. “Agents can operate against governed information, while controls such as access permissions, retention policies and audit trails continue to apply. Individually, these may look like productivity improvements. Collectively, they represent something much bigger: a growing digital workforce capable of acting across the enterprise. And this is where organisations need to think beyond the individual agent.”
However, Hlatshwayo emphasises: “The challenge isn't creating agents. It's governing them. Imagine an organisation with hundreds of agents. One manages content. Another responds to IT incidents. Another supports data governance. Others operate within finance, HR, cyber security, customer service and supply chain processes. They may also come from different technology providers. Without a common approach to governance, organisations risk creating AI sprawl, with multiple autonomous agents accessing information and executing actions across the enterprise.”
He says this scenario raises fundamental questions such as, 'Who authorised the agent?'
What information can it access? What actions can it perform? Which other agents can it interact with? When should a human approve an action? And, critically, can we audit what happened afterwards?
Agentic AI needs a platform, not more agents
Hlatshwayo believes agentic AI ultimately requires a platform approach.
He says: “The platform becomes the control plane for enterprise AI. A platform provides something an individual AI agent cannot provide on its own: a governed operating environment. It can establish the guardrails within which agents operate, providing a framework for identity, permissions, policy, security, monitoring and auditability.”
“Equally important is orchestration,” he says. “A business process rarely happens inside one application. Consider an employee reporting a technology problem. An ITSM agent may help understand the incident. Another capability may need to retrieve supporting knowledge. Another may interact with infrastructure or security services. Once resolved, another agent could help capture new knowledge from the resolution.”
He notes: “The value is therefore not simply the intelligence of each agent. The value is the ability to orchestrate multiple agents, enterprise information, applications and human decisions into a governed business process. From systems of record to systems of action. This is where established enterprise platforms have an interesting advantage. Platforms such as ECM, ITSM and data governance already sit where enterprise information, policy and business processes intersect.”
Hlatshwayo says agentic AI can increasingly turn those environments from systems that simply store and manage information into systems capable of taking governed action. Agents can perform specialised content tasks while inheriting established access controls, governance and audit mechanisms.
“The larger opportunity, however, goes beyond any one vendor or product,” he says.
“As the number of enterprise agents grows, organisations will need to think about AI architecture in much the same way they think about any other critical enterprise capability. The organisations that succeed will likely be those that can give agents enough autonomy to deliver meaningful business outcomes while maintaining the governance, security, auditability and human accountability that enterprises require,” he says.
“Agentic AI may distribute intelligence across the enterprise, but organisations cannot afford to distribute control. That is why the future of enterprise agentic AI is not simply about agents. It is about the platform that governs, orchestrates and connects them,” Hlatshwayo concludes.
To read more about agentic AI, download a free copy of The Agentic Genome here.
Share
