Your AI agents aren't taking over. You're handing over.

In the age of Agentic AI, many organisations are asking: how do we make sure our AI agents don't do things we don't want them to do? How do we keep humans in control? How do we stay ahead of what the technology might do next?

These are entirely reasonable questions. But Pascal Bornet, global AI authority and co-author of The Human-Agent Orchestrator (2026), offers a reframe that changes everything:

Agents do not take control. Humans abdicate it.

With this understanding, we can shift the governance conversation from concern about what AI might do to clarity about what leaders can do. For Australian IT and business leaders navigating the agentic era, it points toward a more practical and empowering approach to staying in control.

How oversight drifts - gradually

The ceding of control over intelligent agents often arises organically from the way organisations operate. An AI agent is deployed to handle a category of routine decisions. It performs well. As confidence in the agent grows, the team's attention naturally moves to other priorities. A few months later, the scope of what the agent handles may have quietly expanded. The operating environment evolved, yet the governance framework didn't keep pace. Team composition shifts. The people who established the original oversight criteria move on to other roles. As a result, the organisation may find that its governance framework no longer fully reflects how the agent is actually operating.

This is what Bornet calls Default Governance: the state in which autonomy expands not through intention but through the natural passage of time and competing demands. It happens in the absence of governance designed to stay current. And understanding it is the first step to getting ahead of it.

Why governance needs a different kind of attention

Bornet's research, conducted across 432 organisations over four years, reveals a consistent pattern: capable agents failed not due to weak technology, but due to a missing management layer. Governance is that management layer, and it is the one that most consistently receives less deliberate design than the agents themselves.

This makes sense in context. The technology is new, exciting, and visible. The governance framework is less visible: it lives in documents, decisions, and review cadences rather than in dashboards and demos. It is easier to focus energy on building and deploying agents than on designing the ongoing oversight systems that keep them well directed over time.

For Australian organisations in regulated industries such as financial services, healthcare, government and legal services, getting ahead of this is particularly valuable. Regulators globally are developing specific frameworks for agentic AI oversight, and the organisations that have established clear governance now will be well positioned as those frameworks take shape.

Three areas worth examining

Drawing on Bornet's research, three dimensions of governance deserve particular attention in any agentic AI deployment.

Documentation. Governance decisions that live primarily in the memories of the people present at launch become harder to maintain as teams evolve and contexts shift. Documenting the key decisions - autonomy levels, escalation criteria, review triggers - creates a foundation that can be actively maintained and handed over as circumstances change. It also makes it much easier to assess, at any point in time, whether the governance framework still reflects how the agent is actually operating.

Ownership. Clear ownership is what keeps governance active rather than passive. For every agentic deployment, there is real value in naming a person or team responsible for the ongoing calibration of oversight: someone who reviews it regularly, updates it as the environment evolves, and ensures the governance framework stays current. Without clear ownership, even well-designed governance tends to drift toward whatever the agent has been doing most recently.

Genuine monitoring. There is an important distinction between nominal oversight and meaningful oversight. Meaningful oversight means humans with the time, training, and clarity to understand what they are looking for, and the authority to act when something needs to change. Designing monitoring that is genuinely actionable, rather than simply present, is one of the most valuable investments an organisation can make in its agentic governance.

Governance as a living capability

One of the most practically useful ideas in Bornet's framework is that governance is not a launch activity - it is a living capability that needs to evolve alongside the agents it oversees. As agents demonstrate reliable performance across a well-defined task set, it is both appropriate and efficient to extend greater autonomy. As context shifts - when agents encounter novel situations, when risk parameters change, when regulatory requirements evolve - governance can move in the other direction. The organisations that approach this well treat their governance frameworks the way they treat other critical operating documents: reviewed regularly, updated deliberately, and owned clearly.

The practical implication is that governance design deserves the same quality of attention as agent design. The two are equally important to the outcomes that agentic AI delivers, and treating them as equally important is what separates organisations that scale agentic AI confidently from those that find themselves managing unexpected consequences.

A more empowering lens

The most valuable shift that Bornet's Default Governance concept offers is not a new tool. It is a new lens.

The productive question is not: how do we stop our agents from doing things we don't want?

Instead, it is: how do we design governance frameworks that stay current, stay owned, and provide genuine, rather than nominal, oversight of the systems we have deployed?

That is ultimately a leadership question. As Bornet puts it in The Human-Agent Orchestrator:

"Old leadership controlled the work. New leadership designs the system that produces it."

Designing that system, including its governance, is one of the most important and most rewarding things leaders can do in the agentic era. It is what keeps the human in the loop, not as a bottleneck, but as the conductor of an increasingly powerful ensemble.

* Who is Pascal Bornet?

Pascal Bornet is not a commentator on AI. He is someone who has spent decades building it, implementing it, and studying its impact at the coalface. Bornet spent over two decades as a senior executive at McKinsey and EY, where he established and led their Intelligent Automation practices and implemented AI and automation initiatives for hundreds of organisations worldwide. He has been consistently ranked among the top 10 global leaders in AI and automation, is a member of the Forbes Technology Council, and has delivered keynotes at over 100 events worldwide each year. He has authored four best-selling books on the subject: The Human-Agent Orchestrator, Agentic Artificial Intelligence, Irreplaceable and Intelligent Automation.

Pascal Bornet will be sharing his insights at an exclusive, invitation-only event in Australia in August, hosted by PhoenixDX.

How PhoenixDX helps organisations build governance that holds

Getting governance right is both a strategic and a practical challenge, and it is precisely where PhoenixDX operates. We work with organisations to design and implement agentic AI governance frameworks that are documented, owned, and built to evolve: defining autonomy levels, establishing escalation criteria, and creating the oversight models that keep humans genuinely in the loop as deployments scale. If you are building out your agentic AI capabilities and want governance that scales with them, we would be glad to help.