Understanding the simple side to Agentic AI

Agentic AI sounds technical, expensive and complicated. But for many businesses, getting started is much simpler than it seems.

In this episode of AI for South Africa, Maxine breaks down agentic AI in plain English and explains why you may not need a massive development project to start using it.

At its simplest, agentic AI means giving AI a job and allowing it to carry out the task rather than simply telling you how to do it. Instead of asking a chatbot how to create a report and then doing the work yourself, an agentic system can follow the process and complete the job.

The easiest way to think about it is as stairs, not a giant leap.

You can start with tools you may already have access to, write down a clear SOP or playbook, give the AI one well-defined job, test what it does and correct it as you go. From there, individual skills can be connected, scheduled and gradually turned into larger automated workflows.
In this episode, we cover:

• What agentic AI actually means
• Why it doesn’t always require custom software
• How an SOP can become an AI skill
• How to train and improve an AI process through feedback
• How individual AI skills can be chained together
• When scheduling turns a workflow into something that runs on its own
• When you can build it yourself and when it makes sense to bring in a developer
• Practical examples for business, HR and everyday life

Maxine also shares examples including an accounting practice automating document follow-ups, an HR onboarding process built from a few connected skills, and a weekly family planning workflow that brings school notices, appointments and activities together automatically.
Your challenge for this week: pick one task you or your team repeat regularly. Write down the steps as if you were training someone new, give that playbook to the AI tool you already use, and run it once.

Don’t worry about scheduling it or building something complicated yet. Test it, correct it and see what happens.

Agentic AI doesn’t have to start with a huge technical project. Sometimes it starts with one repetitive task and a really clear set of instructions.

Subscribe to AI for South Africa for practical conversations about AI, automation and how these tools can actually be used in everyday work and business.

#AIForSouthAfrica #AgenticAI #ArtificialIntelligence #BusinessAI #Automation #AIProductivity #AITools #SouthAfrica

Transcript

Copy Transcript Copied

SPEAKER_01 0:04

Next. This is AI for South Africa.

SPEAKER_02 0:12

Let's go.

SPEAKER_00 0:16

Hello, good night Maxine, and welcome to AI for South Africa. Clients keep coming to me and saying, Maxine, I need to build this. And they've got a whole picture of in their head, like a dev team, months of work, proper software project. And then we sit down and actually talk about what they want. And more often than not, it's it's not actually a huge project. It's it's a skill or a couple of skills that are created that are chained together, doing one thing and then handing it over to the next. And this is happening enough times for me to like I can't ignore it anymore. I can't ignore the pattern. People are picturing something way bigger and complicated than what agentic AI actually takes to get started. I think there's this gap between what agentic AI sounds like and what it actually takes and costs in people's time, money, and honestly, a lot of unnecessary stress. So today we're pulling it apart. What is agentic AI? What does it actually mean in plain language? And do you need a developer and when do you need a developer? Spoiler, it's a lot further than what you think. So let's get started with the word itself. Agentic. Because I think agentic is doing a lot of damage before we've even explained it. Agentic just means that AI acts, it doesn't just answer, but actually go into what the task is. So you would normally go in a chatbot and say, How do I do this report? And then it gives you an answer, and you would copy and paste, and you would do you would do the report yourself. Now with AI GenTech, it goes and does the job for you. That's it. You just ask it, I want this report, and it does it for you. The whole process is done. You give it a job, it goes and it does it. That's it. That's the whole concept underneath this huge word that everybody thinks is a lot of development, a lot of techies sitting in massive data warehouses producing. So where people get stuck is thinking that acting on its own. So AI acting on its own means it needs to be custom built, it needs to be wired into your system, it's coded from scratch, and sometimes it does. I mean, I'm I'm not gonna pretend that that that layer doesn't exist because there's sometimes that it does. But there's this thing I want you to walk away with today. A genetic AI stair. It's not a jump forward, like you know, it's not like this huge hole that you jump into. There isn't this big technical cliff that you have to jump off of. There's a set of steps that most people can climb. And there's quite a few that you can climb on your own that you already have access to. So let's walk these steps, right? Step one is the tool that you're probably already paying for, or it could be without touching a line of code. Things like cowork, chat GPT app, co-pilot in Microsoft 365. These aren't stripped-down demo versions of real AI. They're generally capable of running tasks and not just chatting anymore. So the tool that you're using. Step two is what we're going to call the playbook, or in business terms, an SOP. It sounds simple because it's that simple. It's just the steps that you write down, like for a new employee. When this happens, do this, then this, then this, and then check for this. And this is the perfect outcome. If you can explain how to do a task to a human, you can hand the same playbook or SOP to an AI tool, and it becomes what people are calling a skill. One job clearly defined with the steps on how to follow it. Then step number three. This is where it starts to feel like a little bit of magic, but it's not really. It's just judgment and correction. You don't need to be perfect at it on day one because the more you do it, the better you get. You run it, you watch what it does, and you correct it. When you see this kind of email, don't do that and do this instead. Um, if you see this line of calculations in my Excel, rather do this than that. And then the correction gets folded into the process. It's a conversation, it's less like programming and more like training someone new, you know, taking them through the process. You're not writing code, you're giving feedback. So step A has result B and C. So if B arrives, I want you to do this. If C arrives, I want you to do this. Be the same way as when you were cooking and someone says, okay, I've put the water in the pot. What do I do now? Then you would give them instruction. Four is chaining. And this is where a lot of my clients think that big development projects are actually landing. It's not one skill doing everything, it's a skill that finishes one job and then hands it off to another skill and then another skill, which hands it off to skill number 52. Each one is simple on its own. If you string them together, it can look like something really, really big and massive. And it can orchestrate a full flow. Step number five, the one that people jump to first in their imagination, is scheduling. So set it to run on its own in the morning. You know, every Monday morning I want you to fire this skill, and that skill has a set of instructions to go and find some stuff on the internet. Once it finds that stuff on the internet, I need it to create a Word document for me, which is another skill, and write a report for me in Excel, which is another skill. All of those things are basic SOPs or playbooks that have been created that are triggered by the fact that you scheduled it in the morning. There is a point where you do need technical help. And some people don't want to build this stuff. They just want to be like, Max, I don't have the time to build this, I need it to be in place. You do it, you the experts. 110% agree. If that's what you want, we build it, we put it in place, and you just call it and you just use it. There is that space. But if you're the person who wants to experiment and play, you can so do it. If you need something that requires sensitive documents being processed, if you need it to process things on a legacy system that you have, or if there's certain compliant things you need to do. So let's say a document has to be laid out in a certain way for your compliance to set in from the finance guys. Anything that is out of the ordinary or you genuinely want to scale, then call a developer. No problem. And they can help you. Let's get some real examples going here. A small business owner, Fred, runs a small accounting practice in Durban, and every month he's manually chasing the same documents from the same clients, then formatting it into a summary, and that's not a dev project. That's it's just a playbook that runs. It's just a skill that he has. So when the client emails him a statement, he checks for three documents, flags what's missing, drafts a chase up email, and once everything is done, it builds the summary in a certain format that he can then use. He created the skill, he created the schedule, and it runs by itself every second day. Easy peasy. Then on the corporate side, Taboo and HR of a mid-size company, every new hire triggers the same the same run. There's a welcome email, there's an IT request, there's calendar invite, and then a first week checklist. There's one skill that was created for that, and it gets triggered the moment the person lands in the office. Boom! It triggers the welcome email. Once the welcome email is sent, the IT request is sent. That's also a different skill, and then triggers the third one, which is the first week's checklist. So easy peasy, three skills, and it's sorted for him to do with every new hire. And then from a personal point of view, there's a lady, her name is Sarah. She's got a full-time job with two kids, and she sets up a weekly skill that pulls together the weekly school notices, the activities, the appointments into a simple family agenda on a Sunday night. There's no developer, there's no budget, there's no nothing. It's just a clear playbook for her to prepare her week. Different words, the same underlying moves. Name the repetitive task, write down the steps, hand it over to her skill, and schedule it. It's really simple. And that is a GENTIC AI. These things are running with these tasks on their own. Yes, there's input from us from a building perspective. Yes, there's input from us from a data perspective. But once it's set up and it has the data all there ready to go, it's running and doing it on its own. So here's your homework for the week. I want you to actually do one of these. Pick a task for you or for your team, whatever it is, that at least happens weekly or daily, or that you want some information on. Write down the steps on a piece of paper or in a notebook or wherever, and write it down like you're actually training a person so that you could give this task to somebody else, you know. Um, open whichever tool you're using, ask them to help you create a playbook and run it once and see what happens. If it's wrong, run it again, correct it, have the conversation, go back and forth, correct it and see what happens. That's it. You don't have to schedule anything yet. I don't want you to do anything fancy. I just want you to try and write a playbook and run it in Claude, ChatGPT, or Microsoft and see what the outcome is and see the potential that it could help you. So next time someone tells you that they're doing a Gentech AI, or a Gentec AI needs a massive project and a technical team, you just push back, maybe gently at first, because now you know it's just stairs and it's not a leap and it's not this huge technical thing. It's just a word. Next episode, we're gonna touch on how we can use these elements for more personal productivity, what it takes to actually get productive, and what that setup looks like. I hope you enjoyed today's episode. Please share your tests and actions with me if you do get a gap to have a try and share this with anybody who you think would would want to be interested. My name is Maxine, and this is AI for South Africa.

SPEAKER_01 11:57

Slow breath. Over Cape Town, a disarming, new trees, new waste, a different country.