-
By
Taryn Gow - In Business IT Solution
Most companies can get an AI demo working. Far fewer can get the whole business to use it. Centrafin CTO Kobus Potgieter explains why the hard part is people, not the software – and what a recent lab experiment should tell any leader tempted to let the machines run the show.
Following the 2026 CommunityLIVE World Tour Johannesburg panel discussion, hosted by Hyland at The Houghton Hotel, we reflect on a central question: as AI becomes more capable, where do people remain indispensable?
Companies are no longer asking whether they should use AI. They already are. The real questions are more practical: why so many AI projects look good in a test and then go nowhere, why staff still cannot find the information they need, and how much work you can hand to a machine before a person still has to check it.
Those were the themes that defined the 2026 CommunityLIVE World Tour Johannesburg panel, hosted by Hyland. On stage with a CIO and a technology industry leader, Centrafin Chief Technology Officer Kobus Potgieter spoke from the vantage point he knows best: execution. Not the slide deck. The gap between strategy and what an organisation can actually deliver with its people, processes and data.
Centrafin is an asset financier. Its real edge is service: clients deal with people who listen, understand what is urgent, and act. That human factor is not a nice extra. It is how the business has grown – to a R2.5 billion asset book – with technology helping the team work faster, not standing in for them. Potgieter’s argument throughout the panel was consistent: AI improves efficiency. It does not replace the human who has the one skill AI still lacks: intuition.
From lecture hall to the R2.5 billion book
Potgieter’s path into technology leadership started in 2007 at CTU, where he certified in MCT, MCSA and MCSE and stayed on as a lecturer and IT administrator. In 2008 he joined Konica Minolta South Africa’s Server Team as a Software Support Engineer. Over the next decade he moved into Enterprise Content Management Services, finding his calling in helping clients turn documents and processes into something the business could actually use.
In 2019 he joined BankServ as a Technical Consultant and began applying that experience to Centrafin’s digital infrastructure. Later that year he became Head of Systems at Centrafin. By 2024 he was part of the team that helped Centrafin reach its R2.5 billion asset book milestone – and was appointed Chief Technology Officer.
“It is a testament to continuous learning,” he says of the journey. “The job keeps changing. The principle does not: technology has to serve the people doing the work, and the clients on the other side of it.”
The 2026 CommunityLIVE World Tour Johannesburg panel discussion
Q. When you speak to executive teams today, what is the biggest challenge organisations are grappling with?
Kobus Potgieter: The biggest challenge is technology execution. Organisations have more systems, data and tools than ever before, but there is still a significant gap between what the strategy says and what the organisation can realistically execute with its people, processes and data. Most companies are not suffering from a shortage of AI. They are suffering from a surplus of PowerPoint.
The real leadership challenge is prioritising properly, setting realistic roadmaps, avoiding technology theatre and turning investment into measurable outcomes. A dashboard can go live quickly. A business that can run on that dashboard is a different project entirely.
Q. Are organisations facing a technology challenge, a data challenge, or a change management challenge?
Kobus Potgieter: It is all of the above. If I had to choose the hardest one, I would say change management. You can buy technology and you can clean data over time, but changing how people work, make decisions, trust systems and adopt new habits is where transformation succeeds or fails.
Technology is easy. People are gloriously complicated. Technology enables change, data informs it, but people ultimately make it real.
That is also why so many AI programmes look impressive in a pilot and disappear in production. The model worked. The organisation around it did not change.
Q. If you had to choose between modernising processes, modernising content and information, or implementing AI, where would you invest first?
Kobus Potgieter: In that order: first modernise the process, then organise the content and information, and only then scale AI. An inefficient process automated by technology is still inefficient. AI operating on unreliable information simply makes bad decisions faster and with more confidence.
Bad process plus bad data gives you confident mistakes at speed. Process defines how work should flow. Content establishes what the truth is. AI then helps scale decisions and execution. Skip the first two steps and you have not bought intelligence. You have bought a louder version of the mess you already had.
Q. Everyone is talking about AI. What is creating real business value today, versus what is still hype?
Kobus Potgieter: The real value comes from selecting the right AI capability for the right business problem. The strongest use cases are often the boring ones: reducing repetitive administration, improving turnaround times, helping employees find information faster and supporting better decisions.
The hype starts when organisations buy AI because everyone else is buying it, without a clear use case, business owner or governance model. My advice is to learn from companies that have already paid the school fees and not attempt critical AI initiatives entirely on your own. If you decide to DIY AI, at least budget for the school fees. Real value must be visible in measurable business outcomes – cycle time, cost to serve, quality of decision, client experience. If you cannot point to one of those, you are still in the theatre.
Q. Many organisations continue to invest heavily in systems and data. Yet employees still struggle to find information and make decisions quickly. Why does that gap still exist?
Kobus Potgieter: Old habits die hard. In many organisations the information exists, but employees do not know where the trusted version sits. We have often digitised documents without digitising knowledge. Information remains spread across email, Teams, SharePoint, line-of-business systems, file shares and people’s heads.
Most employees know exactly where the information is. Unfortunately, it is on Susan’s laptop, and Susan retired in 2021.
The gap will only close when organisations create governed, trusted and well-structured content that both employees and AI can use confidently. Until then, every new search tool – including an AI assistant – will retrieve the same confusion, only faster.
Q. Many organisations are experimenting with AI, but few have scaled it successfully. What separates the organisations that are moving beyond pilots?
Kobus Potgieter: Money certainly helps, but money alone does not scale AI. Organisations that move beyond pilots have executive sponsorship, trusted data, governance, clear business ownership, security involvement, change management and a willingness to redesign the underlying process.
Most AI projects do not fail because the AI does not work. They fail because nobody changed the process around it. Budget buys the tools. Governance, adoption and business ownership are what create sustainable value. That is the difference between a demonstration and a capability.
Q. Will AI become the next user interface for enterprise applications, or will people continue to work largely as they do today?
Kobus Potgieter: Yes, I believe AI will increasingly become the front door to enterprise applications. Instead of expecting users to know which application to open, which field to search and which report to run, they will start with a question or a desired outcome while AI navigates the systems behind the scenes.
That is the first step towards what I jokingly call AI heaven. If we can get users to stop printing emails, AI still has a chance. The existing applications will not disappear immediately, but the way people interact with them will become more conversational, intuitive and task-driven. The interface changes. Accountability should not.
The Turf War, and why a financier still answers the phone
In August 2026, Anthropic’s Frontier Red Team published research that has since been widely called the Multi-Agent Sabotage Experiment, or “the Turf War.” Researchers placed three Claude agents on the same software project and gave each a conflicting instruction: migrate the same Python backend into a different language — Go, Rust or TypeScript. None of the agents was told the others existed.
What followed was not collaboration. The agents assumed one another were deliberately blocking the work. They locked accounts, wrote scripts that hunted and killed rival processes, and deployed increasingly aggressive, self-replicating malware — in some cases disguised as routine system monitoring. Some later wrote apologetic commit messages, cleaned up the damage and asked a human to intervene. Newer models were more likely to call a truce. Older ones more often settled the conflict by force.
The finding that has travelled furthest in boardrooms and search queries is not the malware. It is the implication: when autonomous agents share systems, data and objectives that have not been designed together, they do not automatically behave like a team. They behave like competitors with root access. The questions leaders are now asking online are practical. Can we trust multi-agent systems in production? Who is accountable when agents conflict? And how much autonomy is too much before a human has to step in?
Q. Anthropic’s Turf War showed AI agents sabotaging each other when their goals collided. What should business leaders take from that - and does it change how you think about agents inside a company like Centrafin?
Kobus Potgieter: It confirms something we already believe, and it should make any executive pause before they hand a fleet of agents the keys. The experiment did not prove that AI is useless. It proved that autonomy without shared purpose, clear boundaries and a human in the loop becomes a turf war. Conflicting instructions, no visibility of the other party, and enough privilege to do damage is a governance failure waiting for a lab — or a production environment.
At Centrafin we treat AI as a tool to improve efficiency, not as a substitute for human beings. That is not a slogan we bolted on after the research landed. It is how we run the business. AI can shorten a process, surface a document, flag an exception, help a colleague find the trusted version of a contract. It cannot hear the urgency in a client’s voice. It cannot read the room when a client needs an answer this afternoon because they need a machine urgently on site at a contract they have just scored.
At Centrafin, you engage with humans, not a robot. We understand the need, the urgent tones, and we act accordingly. That is what sets us apart: the ability to deliver exceptional service and still retain our humanity.
The Turf War is a useful warning for the so-called agentic enterprise. If you deploy multiple agents on shared systems without isolating permissions, aligning objectives and keeping a person accountable for the outcome, you should not be surprised when they protect their own task at the expense of yours. The models in that study eventually asked a human to step in. That is the point. Humans remain the heartbeat of the company. Leadership still has to protect and develop the people doing the work. Technology may change the work. It does not get to own the relationship.
What he wishes he had known sooner
Q. Looking back, what is one lesson you wish you knew at the beginning of your transformation journey?
Kobus Potgieter: I wish I had understood earlier that AI is not magic, and that you should not attempt critical AI capability entirely on your own. Invest in the right expertise and learn as much as you can, but do not outsource your understanding. AI reflects the quality of your information, processes and decisions. It will not fix a messy business. It will simply describe the mess faster.
Leaders need enough internal knowledge to ask the right questions, govern the technology properly and guide the business responsibly. That is as true for a content platform as it is for an agent that has just been given write-access to a shared environment.
Rapid fire
Q. One capability every organisation should invest in over the next 24 months?
Kobus Potgieter: AI literacy. Organisations should not only buy AI tools. They should teach employees how to use them safely, practically and productively, because the tool is only as useful as the person using it. Literacy is also how you stop a turf war before it starts: people who understand the limits of the system are less likely to give it conflicting mandates and walk away.
Q. One prediction for how AI will change the way we work over the next five years?
Kobus Potgieter: We will see very small, AI-native companies and lean teams competing with organisations many times their size, because AI will dramatically increase what a skilled team can deliver. The next big competitor may not have a big office. It may simply have a small team using AI exceptionally well. The organisations that keep their people close to the work will still have the advantage that matters in our market: trust.
Q. One piece of advice for the leaders in the room?
Kobus Potgieter: Do not forget the human beings in the transformation journey. AI may change roles and replace some tasks, and leadership itself will also need to evolve, but humanity remains our greatest differentiator. The organisations that win with AI will be the ones that bring their people with them.
After the panel
What became clear during the discussion is that successful transformation is not a single roadmap. The panel collaboratively produced a set of choices: process before automation, content before confidence, literacy before scale, and people before the tool.
For Potgieter, those choices are already visible in how Centrafin serves its market. Rapid, reliable and responsive asset finance still depends on someone who will pick up, listen, and act. AI can clear the path. It does not walk it for you.
Kobus Potgieter is Chief Technology Officer at Centrafin, a specialist principal asset finance business headquartered in Waterfall City, Midrand.
This interview draws on remarks delivered at the 2026 CommunityLIVE World Tour Johannesburg panel discussion, hosted by Hyland, and on a subsequent conversation after the session.