Leading with speed and assurance: Governing the agentic AI era

Why governance is becoming the competitive advantage for organisations scaling agentic AI

As organisations move beyond generative AI experimentation and begin deploying autonomous agents across the enterprise, governance is emerging as one of the defining leadership challenges of the AI era.

During a recent HotTopics C-Suite Exchange, held in partnership with Informatica, senior technology leaders came together to explore what it really takes to govern agentic AI at enterprise scale. The discussion revealed an important shift in thinking overall. Governance is no longer viewed solely as a compliance obligation designed to slow innovation. Increasingly, it is becoming the capability that allows organisations to adopt AI with speed, confidence and trust.

From defining what an AI agent really is, to determining who remains accountable when autonomous systems make decisions, the conversation reflected an industry navigating unfamiliar territory. While organisational maturity varied slightly, one message emerged consistently throughout the discussion: the organisations that will scale agentic AI successfully are unlikely to be those moving the fastest but those building the strongest foundations.

Defining agentic AI is proving harder than deploying it

Before organisations can govern agentic AI, many are still wrestling with a more fundamental question: what exactly constitutes an agent?

While participants broadly agreed that agentic AI involves systems capable of planning, reasoning and taking actions with varying degrees of autonomy, there was little consensus around where the boundary actually lies between intelligent workflows automation, and truly autonomous agents.

Some organisations described agentic AI as systems capable of making decisions independently with predefined guardrails. Others viewed it as an umbrella term covering everything from simple productivity assistants to complex multi-agent ecosystems collaborating towards one common objective.

Rather than viewing autonomy as a binary concept, many leaders suggested organisations should think about agentic AI as a spectrum. Some deployments remain tightly constrained, while others increasingly resemble digital colleagues capable of coordinating tasks across multiple systems.

This lack of universal definition creates somewhat practical challenges.

Different regulators define AI differently. Clients are beginning to embed their own AI definitions into commercial contracts. Internal governance teams frequently categorise systems differently from technical delivery teams. The result is that governance frameworks must remain flexible enough to accommodate rapidly evolving technologies without becoming obsolete as soon as they are written.

One of the CSE participants summed up the challenge by observing that organisations cannot afford to wait for perfect definitions before building appropriate governance.

Governance is evolving from a brake into an accelerator

Siddharth Rajagopal, Field CTO at Informatica, argued that governance has historically been viewed as something organisations were forced to implement: “Governance has always been seen as something of who's going to own it. Why should we do it?... Largely it's been seen as a stick.”

The emergence of enterprise AI, however, is changing that perception: “With the AI buzz now it can also be turned into much more of a carrot.”

Rather than treating governance as a separate compliance exercise, Rajagopal argued organisations should instead focus on connecting “the right data, the right people, the right processes, the right technologies, and now the right AI models.”

Several other leaders echoed this changing mindset.

While conversations around AI governance initially centered almost exclusively around privacy, security, and regulatory compliance, organisations are increasingly recognising governance in order as an enabler of business transformation. It provides the confidence to delpy AI more broadly while maintaining visibility, accountability and control.

As one executive noted, organisations no longer want governance simply to minimise downside risk. They increasingly need it to help prioritise investments, demonstrate return on investment, and help enable enterprise-wide AI adoption.

Existing governance is not disappearing, it is expanding

Many participants argued that organisations should resist creating entirely separate governance structures.

Instead, creating disciplines such as information security, data governance, risk management and compliance should evolve to accommodate new forms of AI rather than being replaced altogether.

Several organisations described reviewing existing technology, security and data policies through an AI lens instead of creating entirely new governance models from scratch. For some, recognised frameworks such as ISO 42001 provide valuable structure; not because they solve every governance challenge, but because they offer a consistent way of thinking.

Others highlighted that governance increasingly needs to become risk-based.

Machine-learning models, generative AI systems and autonomous agents each introduce different levels of uncertainty and therefore require different governance approaches. Rather than applying identical controls across every AI deployment, organisations are beginning to tailor oversight according to autonomy, customer impact and organisational risk appetite.

As one participant observed: “We don't want to swap the lure of automation for extra liability.”

Accountability becomes complicated when agents collaborate 

Perhaps the most animated discussion centred around not individual AI agents, but what happens when multiple agents begin interacting with each other.

Several organisations already reported operating hundreds (and in some cases thousands) of agents across their businesses, ranging from internal productivity assistants to customer-facing services and highly specialised “super agents” coordinating the more complex workflows.

As these ecosystems mature, governance questions become considerably harder:

• Who owns an agent developed jointly with a technology partner?
• Who remains accountable when multiple organisations contribute data into a shared workflow?
• How should liability be divided if several interconnected agents collectively produce an incorrect decision?

Participants agree there are few definitive answers today. Instead, organisations are increasingly relying on clear contractual agreements, predefined ownership models and governance committees bringing together legal, technology, security, HR and executive leadership before solutions are deployed.

Human oversight is becoming more contextual

Despite growing excitement around autonomous AI, few organisations described removing humans entirely from decision-making. Instead, many are choosing to adopt contextual oversight models.

Routine, low-risk decisions may increasingly be automated end-to-end, while higher-risk scenarios (particularly those affecting customers, regulated industries or financial outcomes) continue to require human intervention.

This reflects a broader shift away from asking whether humans should remain “in the loop” and towards determining where human judgement genuinely adds value.

Participants from highly regulated sectors described carefully balancing efficiency gains against customer outcomes, ensuring autonomous decisions remain proportionate to their associated risks.

Several also highlighted that governance extends well beyond policy documents. One participant captured this particularly well: “Governance needs to be active, always-on oversight.”

Unlike traditional software, AI systems evolve, models drift, data changes and context changes. Governance therefore becomes a continuous operational discipline rather than a one-off approval process.

Data foundations still matter, but metadata may become just as important

While organisations continue investing heavily in data quality, participants suggested that agentic AI may be forcing leaders to think beyond structured datasets.

Traditional governance teams increasingly need visibility across unstructured information, documentation, retrieval systems, metadata, permissions and observability.

Several organisations described building inventories of AI assets, enriching metadata, improving logging and strengthening monitoring capabilities to maintain visibility over increasingly autonomous systems.

Others questioned whether agentic AI itself could ultimately help automate elements of governance. They asked themselves:

• Could agents classify AI systems against emerging regulation?
• Could they identify governance gaps automatically?
• Could they support compliance teams by maintaining documentation or monitoring risk continuously?

Although most organisations remain in the early stages of exploring these possibilities, many viewed AI-assisted governance as one of the technology's most promising long-term applications.

The biggest challenge may not be technology at all

Perhaps the strongest consensus throughout the discussion concerned people rather than platforms.

While technical capability continues advancing rapidly, leadership maturity is struggling to keep pace.

Several participants highlighted a growing gap between executive ambition and organisational capability.

Business leaders increasingly arrive inspired by demonstrations, vendor presentations and social media examples, but translating those into secure, scalable enterprise deployments remains significantly more complicated.

Rajagopal illustrated this challenge through an analogy that resonated across the discussion. Leaders, he suggested, risk being handed a Ferrari before they've learned how to drive a racing car.

The solution is not slowing innovation unnecessarily, but ensuring organisations possess the governance capabilities to "brake on time."

Education, therefore, emerged as one of the most important leadership priorities.

Participants argued organisations cannot wait until every governance question is answered before developing workforce capability. Instead, learning must happen alongside experimentation, allowing employees to understand both the opportunities and limitations of increasingly autonomous AI.

In partnership with