The AI challenge - balancing governance and innovation

The rise of AI as a powerful business tool has put the spotlight on an old computer system maxim; garbage in-garbage out. No matter how much investment is made in developing the best possible algorithms, the output that an AI system generates is only as good as the data that is used to train the system. And this creates a dilemma for organisations seeking to move their AI projects from proofs of concept into production-ready applications.

How do they make the best possible data accessible to AI platforms when its spread between hyperscalers, SaaS applications on on-prem systems?

"Everyone has a messy data estate," says Leo Brunnick, the Chief Product officer at Cloudera. "Even if you are working towards a single data platform, chances are your data is still spread out across multiple platforms."

Everybody is hybrid, either intentionally or by accident.

Leo recently toured parts of Australia, talking to senior executives about the challenges they face with moving AI from projects to production. He said everybody is struggling with the same thing. The traditional approach has been to move or copy data from its origin to where it is processed. But he advocates a different approach.

"Work with the data where it sits and then add a layer on top which brings the AI to that data whether it's on-prem, in the cloud, or in an air-gapped environment."

When comparing Australian organisations with their US peers, Leo says local companies are more reticent to move forward without some degree of assurance that sensitive information is protected or that an agent taking an action is not properly guard railed. In contrast, he sees American companies move ahead faster as they fear being overtaken by competitors.

Many agentic AI projects are getting stuck at proof of concept says Leo. While there's excitement and optimism for what these projects can acheive, there are very few CEOs and CFOs willing to allow agents that are coded to infer a response and do something with it. He says very few C-suites are willing to lean into doing that until they understand and manage governance.

For Cloudera, the shift towards AI has led to a paradigm shift within the business. The first decade of big data was an era of control with a focus on governance, cost management, and deliberate data handling. The following decade shifted toward convenience, with rapid deployment and cloud‑scale flexibility. AI has ushered in a new era of convergence that demands the merging of control and convenience deployed at a pace that the industry has not faced before. Organisations must re‑assert control through governance, cost monitoring, security, and budget enforcement while still benefitting from cloud‑like ease of AI‑driven services.

IT is the new HR department

With the IT team at the forefront of creating and deploying AI agents, their role has shifted from deploying technology to provisioning digital employees. These new users, expected to soon outnumber humans by 200 to one, will have the same rights and privileges that would normally be assigned during employee onboarding through the HR team.

"A friend of mine runs a department at a software company and was unable to get certain staff positions she wanted," Leo says. "So, she made them with AI agents with names and permissions. I'm now seeing AI agents on company org charts with names roles and responsibilities. And I've seen a couple of forward-looking organisations that assigned a manager of agent employees."

Governance tug-of-war

At a global level, the regulation of AI is pulling in two directions. On one hand, there is the desire to exert control to ensure AI tools are not misused. But, at the other end of the rope, there is the desire to innovate quickly and not be overtaken by competitors. This is playing out nationally as well as commercially.

"The current administration in the U.S. wants to unleash the technology companies and technology products, especially in A.I. in an existential race against China. But they're also simultaneously worried about protecting their own infrastructure. I think it's those two minds battling it out on the regulatory side. And as always, but more so than ever, regulations can't keep up with the technology," Leo says.

Organisations looking to navigate these complex waters lean on their cloud partners to support them. But many have opaque security that gives little insight into how their AI platforms work or are secured. This perhaps explains some of the caution we see, particularly in Australia. And that is further fuelled by geopolitical instability that has put data sovereignty under more attention than ever before.

"I believe AI will become essential for all modern economies to keep pace with the rest of the world. While you can debate which sectors are most affected, AI is already making a difference in defence where it improves decision‑making and operational efficiency. It plays a critical role in predicting severe weather, a growing concern as climate extremes increase. AI is transforming transport, shipping and other logistics, delivering significant efficiency gains. No organisation or developing nation wants to fall too far behind by missing out on the productivity benefits that artificial intelligence can bring."