Leadership reimagined: Governing the agentic AI era

The importance of data governance cannot be overstated. It is critical to ensure that data assets are secure, reliable, of high quality and compliant with the latest regulations. For the C-Suite, good data governance is not only about minimising business risk but also increasingly fundamental to organisational growth, not least given the pressure to embrace (and derive value from) artificial intelligence (AI).

During a recent C-Suite Exchange hosted by HotTopics in partnership with Informatica, CIOs, CTOs and CDOs came together to discuss what the shift from static data to autonomous systems means for data governance and how the C-Suite should react to the transition. Key discussion topics included:

Data governance in the age of agentic AI:

• How agentic AI is fundamentally reshaping what data governance means.
• The critical difference between AI agents that are powerful and those that are genuinely trusted
• Redefining authority, influence and accountability in the age of autonomous AI.
• Practical approaches to building governance that works at agent speed in real time.
• What the next generation of leadership looks like: part data leader, part AI architect and part enterprise risk guardian.

The HotTopics C-Suite Exchange: Turning data into decisions

Senior technology and business leaders across multiple sectors gathered with HotTopics and Informatica to talk about the intersection of data governance and agentic AI, exploring how organisations can build a trusted data foundation, promote real-time observability and establish clear human accountability for increasingly autonomous workflows.

This session was hosted by Doug Drinkwater, an experienced technology journalist and editorial and strategy director at HotTopics, and attended by the various C-Suite members, alongside Steve Holyer, data platform leader at Informatica, a Salesforce company.

What emerged was a vibrant conversation examining the key ways that data governance is changing in the face of agentic AI.

How agentic AI is reshaping what data governance means 

During the opening of this C-Suite Exchange, attendees were asked to grapple with the fundamental question of what data governance means in the age of AI. Notably, the conversation is shifting to focus less on simply governing data, instead prioritising business outcomes, particularly with agents making decisions on our behalf, either semi- or fully autonomously.

"The missing piece is not so much the data,” said one of the attendees. “It's the context that sits on top of the data. We are focusing on observability and evaluation. Taking the data layer, which in our case is quite good, and adding a semantic layer with knowledge graphs on top of that to help form relationships between those things."

Another participant countered that while context is important, it was not the only consideration, given the fact that AI agents are being widely implemented across multiple functions. 

“We need to have a human in the loop. Context is important, but use cases are, too, and that will then determine what kind of data you want to have. The worry for me is that when using the data to teach the LLM about various use cases, you don't want to be teaching with poor-quality data.”

"It’s not just about observability of the stuff you know," continued another attendee. "It’s about detection of the stuff you don’t.”

For executives, agentic AI shifts data governance away from back-office data cleaning toward managing autonomous business outcomes. Given the decision-making potential of AI agents, the potential impact of data has never been greater.

The critical difference between AI agents that are powerful and those that are genuinely trusted

Prioritising the context, business use case and verifiability of data can be difficult, however, when executives are facing heightened pressure to adopt autonomous agents as quickly as possible. 

While new technologies have always shifted the debate surrounding data governance, the speed at which agentic AI is evolving represents a fundamental shift in what data governance means. There may be a temptation to implement the most powerful AI agent, regardless of the required guardrails to protect the organisation, or the veracity of the agent’s output.

Powerful agents may be able to read data at scale and automate tasks, but their reliability is questionable. In fact, research has shown that more powerful generative AI models could even suffer from higher hallucination rates than previous iterations. OpenAI’s o3, for instance, hallucinated 33% of the time on personal knowledge questions, while its successor, o4-mini, had a figure of 48%.

“Think about a Formula One car,” explained Holyer. “Most customers tell me that they invest in brakes to slow the car down. I'm going to say it's to speed the car up. If they've invested in good quality brakes, when they go into the chicane, they will brake later because they have that trust.

“That is what data management is. This is how we're looking to use agentic AI to build trust in your data foundation to make things go faster.”

Holyer identified three data governance lessons organisations have learned as the generative AI revolution has progressed:

1. The initial rush where everyone eagerly wanted to engage with the technology.

2. The sudden fear and anxiety about the risks of deploying it unchecked.

3. The realisation that companies investing heavily in their data management are the ones successfully unlocking true business value.

Understanding these lessons is key to ensuring that AI agents combine power and trust.

Redefining authority, influence and accountability in the age of autonomous AI

In light of agentic AI’s impact on data governance, the entire dynamics of leadership have shifted. The participants agreed that autonomous AI strips away the traditional technical shield they could previously employ. Now organisations can no longer treat AI as an isolated IT problem, and business leaders must assume direct liability for the actions of their automated workforce.

“Ultimately, the agent's not going to get sacked,” said one executive, before quipping that the agent too wouldn’t pay his mortgage.

“It's all going to come back to me. And if what's being piped out there is incorrect, then it’s my neck on the line.”

A major barrier to defining authority in the new era is the gap in AI literacy at the highest levels of leadership. Attendees noted that authority cannot be exercised effectively if the C-Suite doesn't fundamentally understand the tools they are governing.

"An issue you've got is people are scared because they don't know,” continued another roundtable member. “This is particularly true if you've got a C-Suite that has built its career in an analogue world. Either your C-Suite changes – or the people influencing your C-Suite need to change." 

With agents potentially having the power to make critical business decisions, accountability cannot be outsourced to software.

Practical approaches to building governance that works at agent speed in real time

With autonomous agents executing actions in milliseconds, traditional, periodic audits are largely ineffective at guaranteeing good data governance.

The roundtable participants outlined several distinct, real-time strategies to navigate this challenge:

1. Agents Monitoring Agents

Executives are automating the governance loop to keep pace with agent speed. AI layers are constantly evaluating other working agents against corporate benchmarks.

2. Standardising the Software Development Life Cycle (SDLC)

Tracking metrics like tone, accuracy and API calls via automated tools allows teams to benchmark standard behaviour before an agent is launched. In terms of academic scaling and remedial processes, what is required of your data to deliver a lifecycle that is systematic and standardised?

3. Implementing training and clear digital strategies

Establishing mandatory training and outlining organisational expectations regarding your data helps ensure agentic AI tools are aligned with corporate strategy.

4. Applying a FinOps methodology

Understanding return on investment is key. Real-time financial tracking can make sure tokens are not wasted and IT budgets are not exhausted unexpectedly. This is particularly key given AI’s ongoing cost pressures around ‘token economics’.

The practical guidelines offered by attendees of this C-Suite Exchange may not always appear to be aligned with the breakneck speed of agentic AI, but in the long-term, such approaches are likely to be beneficial.

“If you're going to spend a decent amount of time creating an agent, you've got to get that time signed off,” said one attendee. “It might mean that we go a little bit slower, or we're a bit more cautious, but hopefully it means that we get better outcomes at the end."

What the next generation of leadership looks like: part data leader, part AI architect and part enterprise risk guardian

The next generation of leaders can no longer simply act as data stewards obsessing over raw data quality, but instead must contextualise data for AI systems and understand exactly what level of data quality is required for different risk profiles.

“Are you leveraging AI to actually help govern your agents or just falling back on traditional governance processes?” asked one C-Suite member at the discussion.

Static IT dashboards are similarly no longer fit for purpose and businesses must actively design architectures where AI models monitor other models and build systems capable of detecting shadow agents created by end-users or third parties. 

At the conclusion of the HotTopics C-Suite Exchange, the assembled panellists highlighted the necessary transition of data governance from a back-office IT cleanup project into a proactive strategy for automated risk management. Within this ecosystem, enterprise leaders must continue to prize accountability and data quality, even if they must now do so in a dynamic context.

“Individuals want to fly the plane while still building the engine,” Holyer said earlier in the session. To avoid this trap of governance being undermined by a lack of clear safeguards, processes and strategies in the agentic AI era, going slower to eventually go fast is likely to be the best approach for safeguarding and growing the organisation in equal measure. 

In partnership with