The way organisations deploy AI can either build durable trust with employees, regulators and customers or not, writes Kershnee Ballack, technology executive: workforce and workplace enablement for Old Mutual. The important question for CIOs is are we creating the conditions in which our people feel genuinely partnered with AI rather than surveilled, diminished or in quiet competition with it?
Not long ago, I was reviewing the output of one of our AI adoption sessions, part of a structured programme we run across a 23,000-person workforce, when one of our team members said something that stopped me mid-scroll. She wasn’t describing what the tool had produced. She said: “We worked through it together.”
That shift in language, from “the system did it” to “we figured it out”, is small. What it signals is not.
It signals that the organisation has crossed a threshold: AI is no longer a tool being operated. It is a participant being worked with. And for CIOs, that threshold changes almost everything about how we lead.
We have spent years building the business case for AI adoption. We have bought the licences, stood up the platforms, briefed the boards. What many of us have not yet fully done is ask the harder question: how do you lead a workforce that now includes participants who are not human? And underneath that, the question that keeps me up at night more than any other, are we doing right by the people already in the building?
Beyond the tool metaphor
The dominant frame for enterprise AI has been utility. AI as automation. AI as efficiency. AI as a productivity multiplier. These are not wrong, but they are incomplete and increasingly, they are limiting.
When I look at how AI is functioning in my own professional environment, I see something more complex. We are rolling out AI to an initial cohort of employees as part of a structured programme, not as a search enhancement, but as an active participant in meetings, document workflows and decision support. Beyond that, we are in the design phase of purpose-built AI agents: scoped, bounded AI that could handle queries, surface insights and coordinate tasks across complex operational functions. These agents, when they exist, will not just answer questions. They will hold context, escalate appropriately and operate continuously.
That is not a tool. That is a colleague, one with a very specific job description, operating hours that never end and zero tolerance for ambiguity in its instructions.
Globally, this pattern is replicating across industries at pace. According to Microsoft’s 2025 Work Trend Index, drawing on 31,000 workers across 31 countries, 75 percent of knowledge workers are now using AI at work, nearly double the rate of six months prior. Yet the readiness gap is stark: while 67 percent of leaders report familiarity with AI agents, only 40 percent of employees share that confidence – a signal that adoption is outpacing the organisational scaffolding needed to support it.
In South Africa, the picture is layered further by context that is entirely our own. Skills scarcity in specialised tech roles is not an abstraction, it shapes every hiring decision, every capability hub we build, every pipeline programme we run. Infrastructure constraints make resilient digital operations non-negotiable, not aspirational. And in financial services, we carry a particular weight: the organisations in this sector are not peripheral to economic inclusion in this country. We are central to it. Which means that how we deploy AI, who benefits, who is left behind, whether governance is proactive or reactive, is not just a technology decision. It is a social one.
AI is not an experiment here. It is infrastructure. And infrastructure, by definition, must serve people. That is the thread every CIO needs to hold.
What managing an AI workforce actually looks like
AI agents are not going to make most jobs disappear. They are, however, going to change what most jobs require. That distinction matters enormously, and how leaders communicate it will determine whether AI deployments succeed or quietly collapse under the weight of human resistance.
The practical implications of treating AI agents as workforce participants, rather than software configurations, are significant.
First, scope and accountability become real leadership concerns. In designing agent concepts for complex operational functions, one of the earliest and most important conversations is never about capability, it is about boundaries. What decisions can an agent make autonomously? Where does it hand off to a human? Who is accountable when it gets something wrong? These are workforce design questions, not technology questions. The human oversight is not a fallback, it is the design. Leaders who defer these questions to their engineering teams will find themselves managing incidents instead of preventing them.
Second, integration determines value. Many large organisations run human capital management landscapes across multiple platforms, systems that hold some of the most sensitive employee data in the enterprise. When AI agents interact with those systems, whether to answer a payroll query, flag an anomaly, or surface a talent insight, the quality of that interaction depends entirely on how well those platforms are governed, how clean the underlying data is and how clearly the AI’s role within that ecosystem has been defined. The romantic version of AI deployment skips this part. The operational version cannot.
Third, and this is perhaps the most underappreciated dynamic, your human workforce is watching how you treat the introduction of AI. If agents are positioned as efficiency levers designed to reduce headcount, your people will find ways to work around them. If they are framed as surveillance tools or as evidence that leadership has lost confidence in human judgement, trust collapses before adoption begins. The organisations getting this right are the ones investing as much in honest, transparent narrative as in the technology itself. People can handle change. What they cannot handle is uncertainty about where they stand.
Governance is not a constraint. It is the product
There is a version of responsible AI governance that is essentially a compliance exercise. Ethics board. Bias audit. Policy document. Tick. This version satisfies regulators and does almost nothing else.
The version leaders need to build is different, and I want to be specific about what that looks like in practice, because I think our sector is still too vague about it.
In our own AI development process, we have embedded what I call Justice Interrupt Protocols: mandatory checkpoints built into the pipeline before anything goes live, where diverse teams specifically examine power dynamics, historical inequities, and potential for harm, not as an afterthought, not as a legal review, but as a core part of how we build. The name matters. Calling it a “checkpoint” suggests something you pass through. Calling it an interrupt signals that you stop, you examine, and you decide whether to proceed. Those are different disciplines.
The impetus for this came from something I observed during an earlier phase of our AI work, teams that were technically skilled, genuinely motivated, and yet producing solutions that systematically reflected the same exclusions we had been trying to dismantle. Not maliciously. Not consciously. But it was there, embedded in data choices, in assumptions about user behaviour, in which problems were prioritised. Good intentions and great technical skills are not, on their own, sufficient. You need structural interrupts.
South Africa’s regulatory landscape gives us a strong foundation to build on. Popia provides an established framework for responsible data handling, one that applies with equal force when AI systems enter the picture. The Financial Sector Conduct Authority has consistently demonstrated an appetite for innovation within accountability, and its growing attention to algorithmic decision-making reflects exactly the kind of proactive oversight that builds durable public trust. Internationally, the EU AI Act is raising the bar for how technology partners design and deploy systems – and that rising standard benefits every organisation operating in interconnected digital ecosystems.
For CIOs, this creates both a responsibility and a strategic lever. Organisations that build governance in early – that can demonstrate how their AI agents make decisions, where human oversight sits, and what the escalation paths look like – will move faster in the long run, not slower. Regulators and boards trust what they can see. Governance is not a constraint on AI ambition. It is the foundation that makes sustained ambition possible.
The test I return to is not regulatory. It is this: do the people in your organisation feel that AI is working for them, or on them? That answer lives in your governance architecture. It does not live in your communications strategy.
The reskilling imperative – what it actually takes
Every CIO conference includes a panel on reskilling. Most of them cover the same ground: digital literacy programmes, AI awareness campaigns, a few pilot cohorts. These are not bad. They are not sufficient.
What I have learnt from leading a large technology function through a period of significant platform and capability change is that reskilling for an AI-augmented workforce requires three things that most programmes miss.
The first is psychological safety before technical training. People do not engage honestly with new tools when they are afraid those tools are cataloguing their inadequacies or, worse, building the case for their replacement. Before we teach people to use AI, we need to be explicit about what it is for and what it is not for. The conversation about purpose must come before the conversation about features. If your people do not believe you when you say AI is here to support them, no amount of training will produce genuine adoption.
The second is role-level specificity. Generic AI training produces generic AI usage. The question is not “how does AI work?” but “how does AI change what a good human capital business partner does on a Tuesday morning, and what does that free them to do instead?” Role-specific use cases, developed with the people who hold those roles, are far more effective than enterprise-wide rollouts that treat a payroll analyst and a solutions architect as the same learner. The goal is not to train people on AI in the abstract. It is to give them a concrete, credible picture of what their working life looks like when AI is genuinely useful to them.
The third is leadership modelling. In South Africa’s talent landscape, where the competition for skilled technology professionals is fierce and the gap between available skills and organisational need is well-documented, reskilling is also a retention strategy. People stay where they grow. But they take their cues from what leaders visibly demonstrate. If senior technology leaders are not actively using AI, talking openly about what they are learning and acknowledging what they are still figuring out, the message that reaches the team is that AI is for everyone except the people at the top. That is not a message any of us can afford to send.
What CIOs should be asking right now
The strategic questions for 2026 and beyond are not primarily technical. The platforms exist. The capability is real. The questions are human and organisational and in my experience, they are the ones we are most tempted to defer.
Do we have a workforce design framework that includes AI agents as named participants, with defined roles, accountability structures, and performance expectations, and have we communicated that framework honestly to our people? Not in a town hall. In the day-to-day decisions they watch us make.
Is our governance architecture designed for the pace at which AI is actually being deployed? Or are we writing policies faster than we can enforce them and calling that responsible?
Are our reskilling investments targeted enough to produce genuine capability, or broad enough only to produce the appearance of awareness? There is a real difference between an enterprise that has trained 23,000 people on what AI is, and one that has changed what 23,000 people do on a Tuesday morning.
And the question I consider the most important: are we creating the conditions in which our people feel genuinely partnered with AI rather than surveilled by it, diminished by it, or in quiet competition with it?
The organisations that answer these questions well will not just deploy AI more effectively. They will attract better talent, build more durable trust with regulators and customers, and create technology functions that are genuinely equipped for what comes next.
AI is not coming for your people’s jobs. It is coming for the parts of their jobs that are not the best use of human intelligence, freeing people to focus on the work that is. The leader’s job is to make sure the people in the organisation believe that. And believing it requires more than a communication campaign. It requires every structural decision you make about how AI is introduced, governed, and grown to prove it.
AI may be the colleague you didn’t hire. But the culture it enters, and the outcomes it enables, remain entirely yours.
