Prompt‑driven AI cannot deliver the reliability, scale or governance that modern enterprises need. Productivity gains have plateaued, operational risks are rising, and global competitors are already moving into a new phase of AI maturity built around autonomous agents.
Cliff de Wit, chief innovation officer at Accelera Digital Group (ADG), says the industry’s fixation on prompt engineering has become a distraction.
“You can’t build a competitive AI strategy around clever wording. Real transformation comes from AI agents that run processes continuously in the background, supported by humans who step in at the moments that matter,” he states.
This model, known as ‘human-in-the-loop AI’, is rapidly becoming the global standard. It allows organisations to automate complex workflows while retaining control over context, compliance and judgement – areas where AI systems still struggle.
The end of the prompt era
Google Cloud’s AI Agent Trends 2026 Report notes that enterprises are moving away from isolated prompts and towards multi-agent ‘digital assembly lines’ that run entire workflows.
Says De Wit: “Businesses are burning time trying to perfect prompts. The real value comes when employees become orchestrators; people who supervise systems, set objectives and apply judgment.”
Google Cloud’s agent architecture guidance warns that agents struggle in “situations with high ethical stakes” because they lack the “moral compass and judgment needed.”
This is especially relevant in South Africa, where POPIA compliance, credit approvals, healthcare triage and legal reviews require human accountability.
“An agent can execute a workflow, but it cannot understand cultural nuance, regulatory intent or ethical consequence. That’s why intelligent pause points – moments where a human reviews or approves an agent’s action – are essential,” de Wit states.
“South African businesses don’t need more prompting; they need precise, governed automation,” he adds. “The organisations that win will be the ones that build strong relationships between their people and their digital agents.”
Local organisations seeing the best results are embracing a hybrid model where autonomous agents are paired with human orchestrators.
Examples include:
- Banking: Agents manage fraud-detection workflows; humans validate edge-case decisions
- Retail: Agents optimise inventory and logistics; humans intervene when customer experience is at stake
- Public services: Agents streamline case management; humans ensure fairness and constitutional compliance
“Human-in-the-loop isn’t a constraint. It’s a competitive advantage,” de Wit says.
What South African businesses must do next
South African enterprises need to shift their focus from teaching staff how to craft prompts to equipping them to supervise and orchestrate autonomous AI agents.
“This requires redesigning business processes so that agents handle continuous, background workflows rather than isolated tasks triggered by human input,” de Wit states.
At the same time, organisations must embed intelligent pause points into these systems where a human reviews, approves or redirects an agent’s action.
None of this can succeed without strong governance, he comments. “Companies will need multi-layered oversight frameworks that define how agents operate, how decisions are escalated and how risks are monitored.”
Most organisations have a framework for this they can already use. Delegation of Authority frameworks determine how risk management and decision limits are managed. “They can be applied to your agentic workers and help you develop the necessary human oversight rules,” he says.
“It’s not about adding more automation. It’s about building automation that is safe, predictable and accountable, and that only happens when humans remain part of the loop,” de Wit concludes.