The next wave of AI doesn’t need a prompt. It sets its own agenda, takes its own actions and reports back when it’s done. And it’s closer than you think.

Agentic AI is the current buzzword in the world of artificial intelligence, and the growth of AI agents is set to explode in 2026. Although still in its infancy for most enterprise-level work, agentic AI workflows have the possibility of radically transforming organisations, including ours.

As with most buzzwords, agentic AI is often misunderstood – or not understood at all. It’s something industry experts believe will change the fundamental operations of many businesses, or at least some of the functions within them.

Agentic AI, simply put, is a system in which the AI can make decisions and take actions without human supervision. Once it has been set up, it can decide what to do, how to do it and when to do it to reach its objective.

An agentic AI system breaks down complex tasks into steps and acts based on each step. It learns from its environment and draws on internal and external details to make decisions.

The difference between an AI assistant and an AI agent

It is easy to get confused between the different terms for AI tools. AI assistants like Microsoft Copilot, ChatGPT, Claude and Gemini are well used by individuals for a broad range of tasks. These tools wait for you to give it an instruction, but agentic AI doesn’t need you to do that.

Agents use the power of the broader AI assistant but with additional specific instructions, and can sometimes includes a knowledge base. Both assistants and agents require human intervention to tell it what to do to get an outcome, but it has now become easier for employees to build AI agents.

An agentic AI system works whether the user is there or not. It completes end-to-end tasks, is deeply integrated into systems so that it can make decisions on its own and learns from what it has done before.

Think of it in this way: using Microsoft Copilot is like using GPS – it suggests routes, but you still drive. Agentic AI is like a self-driving car – it plans the routes, makes decisions about where to go and gets you to your destination without you giving any input.

Agent in the field

What does that look like in practice? A real-time fraud detection agentic AI workflow might look something like this: the agent picks up a suspicious SIM swop request from Cape Town at 2:47 am. The customer lives in Pretoria and does not typically use their device at that time. The agent assesses the possibility of fraud as high and correlates it with 63 similar attempts across the network. It concludes that an organised fraud campaign is under way and immediately blocks the SIM swop while freezing all other changes on the account. It discovers a large-scale phishing website linked to this incident.

The agent sends multichannel security alerts to the customer as well as to the fraud team manned by human agents for further decisions about legal action and strategic intervention.

This all happens within minutes and without any humans involved.

Numbers to know

  • 40% of job roles in businesses on Forbes’ Global 2000 list of public companies will involve direct interaction with AI systems, according to global marketing intelligence company International Data Corporation (IDC).
  • $3 trillion is how much agentic AI can generate in corporate productivity improvements over the next decade, according to enterprise consultants KPMG.
  • 40% of enterprise apps will be integrated with task-specific AI agents by the end of 2026, predicts business and technology insights company Gartner. This is a huge leap from the less than 5% it was in 2025.

One of those deliberate actions is Code Like a Girl. The programme equips girls and women with coding skills and a route into STEM careers, operating across Mozambique, the Democratic Republic of Congo and, through Vodacom Congo, into Burkina Faso. The results are already tangible.

Vision 2030 sets an even bolder ambition: scaling Code Like a Girl to one million young people. That’s a talent pipeline for the continent’s digital economy, built one line of code at a time.

Then there’s Pouko Pouko in Mozambique, a device-financing initiative that prioritises women aged 18 to 45, particularly entrepreneurs in SMMEs and the informal sector. Since its launch in October 2023, more than 146 000 devices have been sold, with 31% of beneficiaries being women. Half of those who opened savings accounts through the programme are women – because access to a smartphone is also access to financial services, to market information and to economic agency.

Agentic AI at Vodacom

Multiple AI assistants and AI agents have been created or are being tested to improve customer service, enhance productivity and explore new business models. Many of these are still in a partially agentic system or in an AI assistant environment.

Several projects showcase Vodacom’s expanding agentic AI efforts, including:

  • Safaricom’s IoT customer support agent autonomously handles support queries related to Safaricom’s IoT product suite. It is integrated into IoT device management systems for real-time diagnostics. The agent triages and resolves issues without human intervention, where possible, freeing up the specialist support team to focus on complex cases.
  • Vodacom Business in Egypt has a Lead A Day agent that provides leads to an SME based on their requirements of a customer who would be interested in their products or services. It sources suitable company information from various industries, creates and sends emails about the potential lead.
  • Loyalty Concierge Agent, a prototype currently running in South Africa, delivers real-time, fully personalised loyalty experiences across digital and USSD channels. The agent builds its understanding of the customer based on their behaviour, value, churn risk and rewards to offer goals and rewards that speak to who they are, where they are, and how they engage.
  • XNPS, a multi‑agent AI system in the Safaricom Kenya environment, transforms data from across different customer survey feedback into actionable intelligence. It analyses verbatim feedback to uncover key trends, critical issues, urgency levels and AI‑driven recommendations, turning raw customer sentiment into prioritised CX actions.
  • CVM Causal Explainer is an intelligence multi-agent system that explains why customer behaviour changes; not just what happened. It applies causal inference to CVM decisions, unpacking the true drivers behind churn, engagement, uptake, or decline, and translating complex model outputs into clear, business‑ready explanations.

Cybersecurity considerations

AI risk is nothing new, but AI agents have raised the bar for technology teams trying to maintain strong cybersecurity positions.

“In the urgency to adopt agentic AI many organisations risk overlooking a critical cybersecurity challenge: the rise of non-human identities (NHIs), which include API keys, service accounts and authentication tokens,” says Jack Hidary, CEO of SandboxAQ, a subsidiary of Google’s parent company, Alphabet.

“These AI agents interact with tools, APIs, web pages and systems to execute actions on your behalf, not just provide advice. Agentic AIs can spawn NHIs in security blind spots that often receive broad, persistent access to sensitive data and systems without the safeguards typically applied to humans.”

Companies will increasingly have to review their security risks as they drive agentic AI adoption.

Creating new jobs

It will become ever more important for employees to understand how agents work and upskill themselves on what AI agents can do.

According to PWC’s 2026 AI Business Predictions report, new skills will be required as agentic AI becomes embedded in companies. This presents several opportunities for new job creation. Those who understand how it works and what it can do will be in a stronger position in a competitive job market than those who don’t.

The rise of agentic AI will correspond with a rise in various new technical jobs, experts predict, but there are positions for those beyond the engineering space. These include:

  • An AI team manager who oversees an agentic team and ensures the agents do what they’re meant to. They also help to design the original agentic AI system. This is someone with deep knowledge of their environment and context, whether it is in finance, human resources or operations.
  • A human-AI mediation specialist who intervenes when human judgement and autonomous AI decision-making is at odds, to ensure high-level governance and reduced risk.
  • An AI strategist who identifies where agentic workflows can be implemented across the organisation in the most effective way for the highest impact.
  • An AI auditor who monitors the output of autonomous agents to ensure it is accurate, in line with the business requirements, meets legal and other ethical obligations, and is of the level of quality required.

  • Curious about AI in the workplace? In a recent episode of our Tech Talk with Vodacom podcast, explore how embracing new technology and experimenting with AI could transform the way you work. Listen here.

Sources: World Economic Forum, Gartner, KPMG, IDC, PwC