South African organisations are moving beyond chatbots and copilots to autonomous AI agents that make decisions, execute tasks and co-ordinate across business systems.
This creates significant opportunities, but also considerable risk when organisations prioritise agent intelligence over the quality of its operating environment.
This is according to Leslie Price, Head of Engineering at Accelera Digital Group (ADG), who notes that AI agents are only as reliable as the data and context they can access.
“Businesses often connect agentic AI to fragmented databases, disconnected applications and inconsistent repositories. Predictably, agents produce inaccurate outputs and flawed recommendations from incomplete or outdated information,” he says.
What many organisations describe as AI hallucinations are often symptoms of a much deeper problem – poor data architecture.
Price says the challenge will become more pronounced as Agent-to-Agent (A2A) communication and the Model Context Protocol (MCP) enable AI systems to interact with one another and access enterprise knowledge at scale.
"Giving agents more autonomy without fixing underlying data foundations simply means mistakes can happen faster and travel further through the organisation," he says.
Hallucinations are often a data problem
Google Cloud’s The ROI of Gen AI report finds that 74% of enterprises using generative AI achieve ROI within a year, 63% report AI-driven growth, and 84% move a use case from idea to production in under six months.
Yet business value alone does not guarantee successful outcomes.
"Many people assume hallucinations are a model problem. "In reality, they're often a data problem. When agents lack trusted context, they fill informational gaps with probability rather than certainty," says Price.
This aligns with Google Cloud’s ROI of AI-assisted Software Development report, which says AI amplifies existing strengths and weaknesses in systems, processes and data. Fragmented ecosystems therefore automate existing problems faster.
The report identifies AI-accessible internal data, healthy ecosystems, quality documentation and robust platforms as prerequisites. Without them, organisations risk inaccurate outputs, duplicated effort and technical debt.
"As AI agents become more embedded in business processes, data quality becomes a business risk issue, not just an IT concern," says Price.
POPIA raises the stakes
For South African organisations, the implications extend beyond operational efficiency, when considered against the backdrop of the Protection of Personal Information Act (POPIA).
"POPIA changes the conversation entirely. An agent operating on incomplete records, outdated information or poorly governed data isn't just generating poor results. It could create compliance exposure," says Price.
As organisations connect agents to customer, financial and operational systems, governance becomes as important as innovation. Agents need accurate, current and trusted information, not merely controlled data access.
"Before connecting agents through A2A workflows or exposing enterprise knowledge through Model Context Protocol, organisations need a governed context layer and a clear source of truth. Otherwise, they're scaling uncertainty," he adds.
Building the foundation for trusted AI
According to Price, the solution is not to deploy bigger models, but to build cleaner architectures.
"Think of data architecture as the operating system for AI. If that operating system is fragmented, every agent inherits those weaknesses. If it's clean, governed and cloud-native, agents can become reliable contributors to business value," he notes.
For South African CIOs, competitive advantage will come not from deploying the latest model first, but from enabling AI agents to operate effectively.
"As organisations embrace A2A, MCP and agentic AI, success will depend less on the sophistication of the model and more on the trustworthiness of the data beneath it. The organisations that lead won't necessarily have the smartest agents. They'll have the strongest foundations," concludes Price.
For businesses seeking both innovation and POPIA compliance, grounding the machine starts with grounding the data.
Share
ADG
ADG is a global strategic technology firm that helps organisations scale intelligently, efficiently and securely. With teams across Africa, the Middle East and Europe, and recognised as Google Cloud Partner of the Year for Africa (2025 and 2026), the company combines world-class engineering with deep business understanding.
ADG operates through three integrated models: Advisory, Implementation and Managed Services. Each can stand alone, but together they deliver the greatest impact by shaping strategy, executing with precision and sustaining long-term value.