From Automation to Trusted Autonomy - Re-engineering the operating model for the Agentic AI era

Enterprise IT has reached the limits of an operating model built around human decision-making. 

For decades, organizations have invested in automation and AIOps for early incident detection, reducing operational noise and resolving problems faster. These technologies have improved efficiency and resilience.  

But the underlying operating model has remained largely unchanged. 

       An issue occurs. A tool detects it. A person interprets the information, decides what to do and initiates an action. 

That model was built for a world in which systems generated indicators and people provided the judgement. Its architectures, workflows, controls and interfaces were designed to support human operators, not intelligent agents that can interpret context, reason across options and act at machine speed. Simply layering AI onto this human-centred foundation will limit its value and create new points of friction. The AI era requires an operating model designed for collaboration between people and autonomous systems. 

Agentic AI is fundamentally changing this model.  

By introducing reasoning, decision-making and, within controlled boundaries, execution into the operational loop, it enables enterprises to move beyond automating individual tasks towards redesigning how IT operations function. 

The opportunity is therefore much larger than faster detection, diagnosis or remediation. It is about moving enterprise IT from reacting faster to understanding and responding better. 

This shift raises a consequential question for senior IT leaders: when AI can reason and act, how much authority should it be given, under what conditions and with whom does accountability ultimately rest?  

The answer will determine whether agentic AI remains a collection of promising pilots or becomes a trusted operational capability at enterprise scale. 

Autonomy is not a technology feature 

The journey towards autonomous operations is often described as a technology progression, but the more important distinction lies in the level of judgement being delegated.  

Traditional automation follows prescribed instructions. AIOps add intelligence by recognizing patterns, correlating related events and reducing operational noise. Agentic AI goes further by evaluating context, reasoning across possible causes, planning a response and, where authorised, executing an action. 

Consider a sudden spike in cloud consumption at 2 a.m.  

In a conventional environment, an alert may wake an on-call engineer, who must determine whether the increase was caused by legitimate demand, a scheduled workload, a faulty release or a cyberattack. In an agentic model, AI could assess operational history and context, then allow the workload to continue, initiate a pre-approved response or escalate the issue to the appropriate person. 

Yet agentic capability alone does not create autonomy. Trust and governance also play an equally important role. A practical equation captures the difference:  

                       Autonomy = Agentic AI + Earned Trust, Scoped and Governed 

Agentic AI is effectively autonomy on a leash. The enterprise begins with a narrow boundary and limited blast radius, then extends authority as the agent demonstrates reliable performance. 

Trust is the real constraint 

Technology can already do more than many organisations are prepared to permit.  

The principal barrier is trust.  

Permanently limiting AI to recommendations constrains its value, while granting extensive execution authority from the outset introduces unacceptable exposure. The practical path sits between these extremes. 

Enterprises can use AI broadly to recommend actions while initially permitting execution only in low-risk, reversible, policy-approved and well-understood scenarios.  

Greater authority should follow demonstrated performance, not technological potential. 

Governance is an accelerator 

Governance is often perceived as a brake on adoption.  

In practice, weak governance can slow progress more decisively. If an agent takes an action but cannot explain why, confidence deteriorates, approvals multiply and people begin to second-guess every decision. 

Effective governance establishes the conditions under which autonomy can expand safely. Policies must be defined before agents are allowed to act. Authority should increase progressively.  

Decisions must also be explained: the agent should show what information it considered, which policy it applied, why it chose an action, how it executed it and what outcome followed. 

When these foundations are in place, governance does not constrain autonomy. It gives leaders the confidence to extend it. 

Context separates pilots from production 

Many AI initiatives succeed in controlled demonstrations but struggle when exposed to enterprise complexity.  

IT estates shaped by acquisitions, technology transitions and years of local decision-making contain inconsistent tagging standards, distinct escalation paths, unique change windows and varying levels of risk tolerance.  

Much of this knowledge is implicit and may not be adequately represented in formal systems. Without that context, an agent may make a technically valid but operationally inappropriate decision.  

A disciplined embed, calibrate and expand approach can bridge the gap.  

• First, capture the organization’s business and IT context.  
• Next, test agents against real tickets, change windows, escalation paths and operating scenarios.  
• Finally, expanding autonomy only after performance has been demonstrated in the enterprise's own environment. 

Deploying a platform or a collection of agents does not create autonomous operations. Technology provides capability. Context makes that capability relevant and dependable. 

Humans remain accountable, but the loop changes 

Autonomous operations do not remove people from IT. They change where human judgement is applied.  

Initially, people may review individual actions proposed by agents. As confidence grows, responsibility shifts towards defining policies, establishing risk boundaries, governing decision systems and overseeing exceptions. 

Oversight should be proportional to consequence. High-risk actions require greater human involvement, while low-risk and reversible actions can operate with less intervention.  

Humans remain in the loop but move from making every routine decision to governing the environment in which trusted decisions are made. 

The leadership imperative 

Agentic AI should not be managed as another tool deployment. Agentic AI introduces a more consequential question: which decisions should remain with people, which can be delegated and under what conditions? Answering it requires three leadership shifts:  

1. From technology adoption to operating-model transformation 
2. From experimentation to operational accountability 
3. From permanent control to progressively earned trust 

Data quality, governance, process redesign and organisational change may be less visible than the latest AI model, but they determine whether experimentation becomes an enterprise capability. 

AI agents will become widely available. Sustainable advantage will come from an organisation's ability to make them dependable in its own environment, grounded in context, bounded by policy, capable of explaining their actions and overseen in proportion to risk. 

What CIOs and IT leaders should do now 

Turning trusted autonomy into an operating capability requires leaders to act deliberately in five areas: 

• Redesign the operating model, not just the technology stack 

Identify where reasoning and execution can be delegated, and redefine the roles, controls and accountabilities around those decisions. 

• Start with bounded, reversible use cases 

Prioritize actions with low operational risk, clear policies and measurable outcomes before expanding the agent’s authority. 

• Build context into the system 

Connecting agents to the operational history, business rules, escalation paths and change windows requires making decisions appropriate to the enterprise. 

• Make explainability and auditability non-negotiable 

Require every material action to show the evidence considered, policy applied, decision taken and outcome achieved. 

• Create a mechanism for earned autonomy

Review agent performance regularly and extend, restrict or withdraw authority according to demonstrated reliability and risk. 

The strategic question is no longer whether AI can act. It is whether the enterprise is ready to let it act. The next era of IT operations will belong not to those that automate the most, but to those that know what to delegate, how to govern it and when to extend the leash.