AI. Autonomous AI agents holds the potential to revolutionise core banking operations, enhance customer experiences, and fortify financial security, it also carries risks particularly in places like Africa, with vast unbanked populations, diverse languages, and varied regulatory frameworks across 54 countries, the writer says.
Image: File Image: IOL
By Prejlin Naidoo
Many people consider ChatGPT the first major milestone of generative artificial intelligence but the technology at the frontier of innovation today is something known as the AI agent.
These goal-oriented AI applications use large language models (LLMs), reasoning approaches, access to tools, and autonomous goal-seeking to respond to prompts and solve problems, while learning and evolving over time.
Anyone familiar with the saga of Truth Terminal the meme-driven AI agent that managed to autonomously raise $300 million in five days simply by asking for funding to upgrade itself — recognizes the immense promise this technology brings.
But while the advent of smart, autonomous AI agents holds the potential to revolutionise core banking operations, enhance customer experiences, and fortify financial security, it also carries risks particularly in places like Africa, with vast unbanked populations, diverse languages, and varied regulatory frameworks across 54 countries.
Financial institutions eager to deploy this promising technology need to do so with care.
Agentic AI has the power to reshape banking by functioning as an independent decision-maker rather than an analytical tool. These systems are able to comprehend complex objectives and deconstruct them into manageable components.
Unlike traditional AI that follows predetermined paths, these agents determine optimal actions to achieve specific outcomes through sophisticated planning and prioritisation.
Building upon the foundations established by frontier models like GPT-4o and Deep Seek 3, agentic AI excels at complex reasoning, analysing problems through multi-step processes, identifying critical elements, and developing logical action sequences to address challenges effectively.
Essentially, agentic AI thinks more like a human, breaking down problems, planning multiple steps ahead, and devising the best approach to reach its goals. So, what separates agentic AI is its ability to execute actions without human oversight. These systems access tools, applications, and organisational systems directly, making informed decisions and taking concrete actions rather than simply offering recommendations.
They establish objectives and determine implementation strategies independently.
These agents evolve through experience, refining their understanding and problem-solving approaches based on outcomes, creating a virtuous cycle of improvement where each interaction enhances future performance.
While conventional AI systems follow rigid programming, agentic AI actively evaluates customer needs in real-time. For example, during an interaction about financial challenges, an AI agent might recognise underlying concerns about retirement security, analyse the customer's financial position, and immediately offer tailored investment products specifically designed for their situation—all without human intervention.
These systems can enhance financial security by analysing vast datasets in real-time, detecting anomalies, and preventing fraud.
Agentic AI can also spawn subordinate agents to handle specialised subtasks, creating a network of problem solvers. A primary financial planning agent might create dedicated sub-agents for market analysis, tax optimisation, and estate planning, then synthesise their findings into comprehensive recommendations—something traditional AI systems cannot accomplish.
While this is exciting territory, the adoption of AI agents in banking also comes with significant risks, particularly as they require extensive access to tools and data to realise their potential.
Though the story of Truth Terminal highlights the potential of AI in financial innovation, it also underscores the risks of giving AI agents too much autonomy.
If an AI agent can independently influence markets, what happens when similar agents are integrated into banking systems? The scale of risk increases exponentially, with potential vulnerabilities ranging from data loss to financial instability.
The challenge is that for AI agents to become truly transformative, they need access to a range of tools and systems across the bank. As with the integration of any new technologies into legacy systems, granting this access introduces new vulnerabilities.
Trust is foundational to the banking sector, and when security is breached, financial institutions suffer reputational harm. When AI agents are tasked with managing millions of transactions and processes, every decision must align with complex business rules and industry regulations. Any lapse could result in data breaches, operational disruptions, or regulatory non-compliance.
At the application level, these risks are magnified considerably, requiring increased scrutiny and more robust governance frameworks. When AI systems operate at the customer interface level, making decisions that affect individual financial wellbeing, the potential for harm escalates dramatically.
Heightened efforts must be made to identify and remove biases within these AI systems. Without proper oversight, an AI agent could make decisions that appear helpful but are actually predatory.
For example, an AI agent may autonomously identify that a customer has overspent, then offer that customer a short term loan to help them cover that shortfall. It’s vital to ensure that these interventions aren’t pushing unnecessary and unaffordable debt onto vulnerable customers.
African banks, while not yet ready to fully deploy AI agents at scale, are currently experimenting with integrating this new technology into their operations. These systems should be designed to address Africa’s unique challenges, from serving remote rural communities to creating financial products tailored to local economic patterns.
Agentic AI presents a watershed opportunity for African banks to enhance efficiency, extend financial inclusion to previously underserved communities, mitigate risks unique to emerging markets, and deliver personalised services that respect local customs and needs.
However, with this transformative technology on the horizon, it’s increasingly vital to navigate the landscape with a discerning eye, embracing the promise while remaining vigilant against the inherent risks accompanying such innovation.
Prejlin Naidoo, CMT Partner at Oliver Wyman
Image: Supplied.
BUSINESS REPORT