Handing a Formula 1 race car to a novice driver does not increase their chances of winning the Grand Prix. In fact, it might well make them dangerous.
By President Ntuli, MD of HPE South Africa
That is the paradox facing South African businesses as they move beyond experimenting with AI and begin building truly agentic enterprises: the more powerful the tool, the more disciplined the operator must be.
While most local organisations remain consumers of AI, the first signs of the agentic enterprise are already emerging. Almost half of South African banks are each investing more than R30 million in AI this year, much of it directed towards autonomous agents. And this is only the beginning. The local enterprise agentic AI market is expected to grow at close to 50% annually through to 2030.
AI is no longer just helping people get work done. It is starting to shape how work gets done, making recommendations, initiating actions, and influencing decisions across the business. The opportunity is enormous. And so are the risks.
In many ways, the agentic enterprise will be the great amplifier, accelerating what works and exposing what does not.
From using AI to owning intelligence
Most organisations today use AI tools built elsewhere to drive productivity. The agentic enterprise flips that model. It enables organisations to build intelligence around their own data, systems, and expertise, turning AI from something they consume into something they own.
What we see globally is an operational shift, where organisations are moving from static workflows, where systems support decisions, to dynamic environments where AI observes, reasons, and acts alongside teams. Intelligence becomes embedded across applications and infrastructure, enabling faster decisions, closer to where they matter. However, as AI becomes more accessible, the real advantage will come from the intelligence organisations create from their own data.
The upside is clear, but the risks are catching up
If the opportunity is producing intelligence, then the challenge is controlling it.
As intelligence becomes distributed, the objective changes. It is no longer just about generating insight, it is about coordinating, governing, and operating intelligence across a constantly shifting environment.
AI agents are not passive tools. They are autonomous, high-privilege actors capable of accessing systems and taking action. Every ungoverned agent expands the enterprise attack surface. In a Dark Reading poll in January, nearly half of the cybersecurity publication’s readers predicted that agentic AI would become the top attack vector for cybercriminals and nation-state threats.
Then, there is the growing risk that AI spending outpaces AI value. PwC found that just the top 20% of organisations globally capture 74% of AI-driven returns. As investment in agentic AI rises, the gap between organisations that successfully operationalise AI and those that struggle to realise meaningful outcomes is widening.
Often, the challenge is not the technology itself, but the absence of the policies and internal guidelines, the knowledge base, operating models and the processes needed to securely translate experimentation into sustainable business value. In this context, data sovereignty and regulatory clarity carry real weight, making governance not a compliance checkbox but what makes AI viable at scale.
Control becomes the differentiator
As AI agents become embedded across operations, governance must move from policy to practice. Organisations need clear controls and processes to regulate who can access AI systems, what those systems can do and how they are monitored and managed.
Simultaneously, security must be built in from the start. Organisations need the ability to prevent incidents as far as possible and to recover quickly when something goes wrong. In a world of autonomous systems operating at speed, resilience becomes as important as protection.
At the same time, we see the conversation around AI economics shifting, from model performance to utilisation, efficiency, and cost control. Organisations need visibility across the infrastructure to better understand what AI can do, what it costs, where it is being used, and whether it is delivering value.
Data is no longer an input. It is the system
Ultimately, the question of AI value comes down to data. Agentic AI does not simply use data differently from traditional AI; it depends on data in a fundamentally more continuous and operational way.
Traditional AI systems typically access data to answer a query or generate a response. AI agents, by contrast, operate across entire workflows. To achieve a goal, they must continuously discover information, analyse context, reason about next steps, coordinate across systems, and take action. Data is therefore no longer a static input. It becomes the operating environment in which intelligence functions.
This raises two distinct but equally important requirements. The first is data infrastructure: the ability to connect, govern and access data wherever it resides. Agents cannot operate effectively if enterprise data remains fragmented across applications, clouds and business units. They need a unified, discoverable and governed view of information that allows them to securely access and act on data across the organisation.
The second is data quality. Accessibility alone is not enough. Once agents can access enterprise data at scale, the accuracy, completeness and reliability of that data become critical. Poor-quality data does not simply produce poor insights; it leads agents to make poor decisions, automate flawed processes and amplify errors at scale. The consequence is not isolated mistakes, but systemic ones.
This is why data readiness has become one of the defining challenges of the agentic era. Organisations need both the infrastructure to bring together distributed data and the governance to ensure that data can be trusted. The organisations that succeed will be those that strengthen these foundations before they scale AI. Increasingly, that will mean bringing AI to the data, rather than moving sensitive enterprise data to AI.
The network becomes the backbone of intelligence
With the agentic enterprise depending on intelligence flowing seamlessly between people, systems, data, and AI agents, the network has become one of the most critical parts of IT infrastructure. Connectivity is no longer a background function. It is a strategic enabler of AI performance, security, and scale.
This is why “networking for AI” and “AI for networking” are becoming central to enterprise strategy. Organisations need networks engineered to meet AI’s performance demands and AI capabilities that can manage the growing complexity of those networks. AI-native networks can simplify deployment, automate troubleshooting, strengthen security, and reduce operational overhead. Built for AI, they provide high-speed,low-latency architectures, reliable data movement and compliance by design, ensuring AI workloads can run efficiently and securely at scale.
The organisations that pull ahead will not be the fastest to deploy agents, but the ones disciplined enough to govern them. For South African businesses operating under real constraints such as infrastructure gaps, regulatory uncertainty, and skills shortages, that discipline is no longer optional. It is the only credible path to making this technology work at scale and the window to build those foundations is narrower than most boards currently appreciate.