What exactly is agentic AI? A plain-English guide for South African business owners

The term "agentic AI" is everywhere right now. Vendors throw it around. LinkedIn influencers can't stop posting about it. Half the software tools you already pay for are slapping the label on their next update.

But nobody is explaining it in a way that helps a business owner in Johannesburg or Pretoria make a decision. So let me do that.

I'm Timo van Deventer. I've spent 18 years building automation, from financial markets to marketing systems at scale, then seven years as a COO wiring up the operations behind real businesses. I write the code, architect the systems, and sit on every build at Aivolution. This is what I think you actually need to know.

The simplest definition of agentic AI

Agentic AI is artificial intelligence that can take a goal, break it into steps, decide which steps to take, use tools along the way, and adjust its plan when something goes wrong. All with minimal human input.

Think of it like this. Regular AI (including ChatGPT) waits for you to ask a question, then gives you an answer. You prompt, it responds. That's it.

Agentic AI doesn't wait. You give it an objective, like "find me three suppliers for X, compare pricing, and draft an order for the cheapest one." It then figures out how to do that. It searches, reads results, calls APIs, re-plans if a supplier's site is down, and comes back with a completed task.

The key distinction: agency. The AI acts on your behalf across multiple steps, making decisions at each stage.

Is ChatGPT agentic AI?

This is the most common question I hear. The short answer: not really.

ChatGPT is a large language model. It generates text in response to a prompt. OpenAI has bolted on features that move it in an agentic direction, like web browsing, code execution, and plugins. But the core experience is still prompt-response. You tell it what to do step by step. It doesn't autonomously decide to go research something, call an API, revise its approach, and deliver a finished outcome without you steering.

Some of OpenAI's newer products and research demos do show genuinely agentic behaviour. But what most people use day to day is still a very capable text generator, not an autonomous agent.

And here's my blunt take: only using ChatGPT is not leveraging AI for your business. It's useful, sure. But it's like using Excel only to make lists. The real value comes from connecting AI into your workflows so it acts on data, triggers actions, and removes repetitive work.

What's the difference between generative AI and agentic AI?

Generative AI creates content. Text, images, code, audio. You give it input, it gives you output. One turn.

Agentic AI uses generative AI as one tool among many. It might generate an email draft, but it also decides when to send it, to whom, based on what data, and what to do if the recipient replies with a question.

Generative AI is the engine. Agentic AI is the driver, the navigator, and the mechanic rolled into one.

What about "AI agent" vs "agentic AI"?

An AI agent is a single entity that can act autonomously. Agentic AI is the broader concept, the architecture where one or more agents work together, use tools, maintain memory across sessions, and coordinate to complete complex tasks.

Think of an AI agent as one employee. Agentic AI is the operating model where multiple employees coordinate, each with their own specialty, passing work between them.

So why am I telling you not to use it yet?

Because agentic AI is not ready for South African SMEs. Not in 2026. Not at the level most businesses operate.

Here's why I say that:

It's expensive to get right. True agentic systems need careful orchestration, error handling, fallback logic, and monitoring. The cost of building one that works reliably is significantly higher than a well-designed workflow automation.

It's unpredictable. When an agent decides its own steps, it can go sideways. In a large enterprise with dedicated AI teams and testing environments, that's manageable. In an SME where one wrong invoice or one rogue WhatsApp message damages a client relationship, it's a risk you don't need.

It fights compliance. POPIA requires that decisions with legal or material effect on a person have human involvement (s71). Fully autonomous agents making decisions about customers, employees, or finances is a compliance problem we won't touch. At Aivolution, we never build fully autonomous decisions with legal or material effect. That's a hard line.

You don't need it. The vast majority of automation value for SMEs comes from structured, predictable workflows with clear guardrails. Not from giving an AI the keys and hoping it figures things out.

What you should build instead

No-code workflow automation with guardrails. That's the answer for most SA businesses right now.

We build these using tools like n8n, Airtable, WhatsApp Business API, and language models from Claude, OpenAI, or Cohere. The AI handles the thinking, like reading an email, classifying a lead, drafting a response, or extracting invoice data. But the workflow is fixed. Step A leads to Step B leads to Step C. The AI operates within boundaries we set.

Human-in-the-loop is part of the design, but it doesn't mean babysitting. The human completes only necessary tasks: a confirm, an approval, a judgment call. Then the system runs. You're not shadowing it. You're not watching every message. You check a dashboard, approve a batch, handle the exceptions. The rest flows.

This is how we build WhatsApp automations that handle customer queries in English and Afrikaans, qualify leads, route conversations, and only pull a human in when the conversation genuinely needs one. Same with AI-powered customer support that handles the repetitive questions and escalates the real problems.

The result: you get the practical benefit everyone's promising with "agentic AI" but with systems that are predictable, auditable, and compliant.

The cost reality

Let me ground this in numbers I work with every day.

A SaaS tool might cost R500/month. Over three years that's around R18,000. Fine if it genuinely solves your problem. If your automatable workload is below roughly R8,000-R10,000 a month in equivalent labour cost, buy the tool. We'll tell you that straight, even though it means we lose the deal.

Hiring someone to do that work costs R15,000-R25,000 a month. That's R216,000+ per year, forever, plus roughly R30,000 in recruitment costs that reset every time they leave.

A custom build starts from R75,000 once-off with a retainer around R2,000/month. Over three years that's approximately R147,000. It runs 24/7, scales without headcount, and replaces the automatable 60-80% of a role. The human you already have gets freed up for the work that actually needs a brain and a relationship.

That's the real build vs buy calculation most businesses never do properly.

Where agentic AI is heading

I'm not saying agentic AI is vapourware. It's real, it's improving fast, and within a few years the tooling will mature to the point where SMEs can use it safely.

When that happens, the businesses that already have structured automation in place will adopt it fastest. They'll have clean data, working integrations, and the operational discipline to add agentic capabilities on top.

The businesses starting from zero will be years behind.

So the play is simple: build the foundation now with reliable workflow automation. Add agentic capabilities later when the guardrails, cost, and compliance picture makes sense for your size.

The bottom line

Agentic AI is a real and meaningful evolution. It's also overhyped, undersupported for SMEs, and not something you should be buying from a vendor who can't explain it in one sentence.

What you should be doing in 2026 is automating the repetitive, soul-crushing work that eats your team's time. Not with a magic autonomous agent, but with well-built workflows that use AI where it adds value and keep humans where they matter.

AI doesn't strip humanity from your business. It removes the tasks nobody should be doing manually so your people can spend time on real human contact.

If you want to see what's actually automatable in your business, book a free 45-minute audit. No obligation, no pitch deck. I'll look at your operations and tell you what's worth building, what's worth buying off the shelf, and what to leave alone.