Let me tell you a story.

One morning, I spent 5 grueling hours reconciling school fees using spreadsheets and bank statements. I did not even know it was 5 hours until I looked up and it was past 1 pm. I was tired, annoyed, and frustrated, and right then and there I knew I was never doing that again. No way. That morning is why I am writing this post.

Last week we talked about whether Nigerian businesses need AI tools or AI agents. The short version is that with a tool you tell AI what to do and how to do it. But with an agent you give it a goal and let it figure out how. If you missed it, you can read the post here.

This week I want to show you what an AI tool and an AI agent look like in real life, using my own work as the case study. Whether you run a business, work in one, or you are a student curious about building your first thing with AI, the pattern is the same.

5 Hours, Manual Labour
Before AI, here is what reconciling school fees looked like for me.

Step One, download every bank statement, dump it into a master spreadsheet, and normalize the new transactions to match the existing format. Two to three hours, easy.

Step Two, the receipts. My staff write receipts by hand as students bring in bank tellers and handwritten notes from parents and guardians. At CBT and exam time we can have 100 receipts to write and sign in a single day. I sit with the stack, look up each payment in the master spreadsheet, and only sign if the transaction is actually there. If it is not in the spreadsheet, either the payment has not dropped yet or it has not been paid. This takes me one to two hours.

Step Three, pivot tables. Who paid, who is owing, who had a discount, who paid for something other than school fees. I had to remember all of it. Another one to two hours.

As tedious as this was, this system was much better than what the bursar had been doing with paper and pencil before I came on board. But the room for human error was still wide and the time it ate from my week was ridiculous.

So that morning I closed the spreadsheet, opened Gemini, and started talking.

How We Built An AI Tool
I studied Computer Science and worked as a programmer for many years, but the way I build with AI now has nothing to do with code, at least not at the start.

I start with a conversation. I tell AI what I want, I ask it how it would design it, and I tell it not to give me any code yet. Let’s discuss. We go back and forth until I feel it understands the problem and I like how it is designing the solution. Only then do we start building.

This way I am not running broken code, fixing it, prompting again, running it again, and watching my afternoon disappear into a prompting loop. The conversation does the heavy lifting before a single line of code is written.

So I started talking to Gemini about building a financial reconciliation system for our school, and one to two days later I had a working prototype of what we now call our School Financial System. Within a week I was using it with the master spreadsheet as backup. Within a month I had moved over completely and only opened the spreadsheet for very old transactions.

Since then we have added functionality the spreadsheet never had, like family tagging, so when a payment comes in under one child’s name the system reminds me there are two siblings to share with. We have better tracking across payment categories, and cleaner summaries without mental gymnastics.

What used to take hours now takes minutes. The first time my staff watched me do it she said it didn’t even take any time at all. When I showed her the system her eyes grew big and all she could mouth was, Wow.

One thing we did not automate is the handwriting of receipts. We could. We chose not to. Our students and parents need that human point of contact when they hand over a payment, and me signing each receipt is part of how I stay close to what is happening in our finances. That is our human in the loop verification step, and it is deliberate.

Then We Built An Agent
The School Financial System is a tool. We told AI exactly what to do and how to do it. To really understand agents, we needed to build one, so I asked Claude what simple agent we could build to demonstrate the idea. Of all its suggestions I picked a travel agent. It was simple and quick to build. Or so Claude told me.

It built the core functionality easily enough. You tell it you want to fly from Lagos to Abuja on Friday. It checks airline availability and brings back every airline running that route on that day. The agent is free and easy to use. Just type your route, date and time and click search. You can try it here.

In building the AI Travel Agent, I learnt that Nigerian Airline APIs are not publicly available so you have to use free sources and scrape the airline websites for comparison. One challenge with this is if the website URL changes, unlikely, or the page address changes, more likely, our agent will fail silently. That’s why we built a watcher agent for our agent to periodically confirm airline flight routes and airports. We also added a Know Your Rights page for stranded passengers dealing with delays and cancellations, because that is the kind of thing a Nigerian traveller actually needs and almost never has at hand.

The next agent I am thinking of is Summary Analysis agent for our School Financial System. I will not tell it what to show me. I will give it the data and ask it what it sees. That is the part of AI I find most powerful, not just AI doing the work, but pointing out the things I would never have thought to look for.

What Does It Cost?
Fair question. Because Nigerians make decisions in Naira, here is the honest answer. Costs run from little or nothing all the way up to around 100,000 Naira a month for building and running serious multistep tools and agents.

You can genuinely start at zero. Gemini has a generous free tier. A small tool can live inside a Google Sheet with a bit of Apps Script. A simple app can sit on the free tiers of platforms like Vercel, Netlify, or Render. If all you want is to think out loud with AI and have it help you solve a problem, you do not need to spend a kobo.

Once you start hosting real tools that other people use, you climb. A paid AI subscription, a paid hosting plan, light API usage when your tool calls AI in the background. That stack stays modest until you are doing heavier work.

I spend more than the average person needs to spend, because I am building daily and I cannot afford to hit usage limits in the middle of a build. You do not need to start where I am. Almost no one should.

The point is not the exact figure. The point is that the entry price for building something useful is far lower than most people assume, and you can grow your spend only when the work justifies it.

So Which One Do You Need?
If you know exactly what you want done and how, build a tool. If you want AI to go and figure something out and tell you what matters, build an agent. Most of the time, what you actually want is a blend. You bring the context and the constraints, AI brings the suggestions and the work, and somewhere in the middle the real thing gets built.

If you are not sure what you need right now, just have a conversation with AI about your pain points. I won’t be surprised if AI suggests something you haven’t even considered.

Start Here
If you take one thing from this post, take this. Pick the task in your week that drains the most hours and gives you the least joy. For a business owner that might be customer service, reconciliation, payroll, or inventory. For an employee it might be report writing or weekly summaries. For a student it might be revision tracking, WAEC practice, or organizing notes across subjects.

Open Claude, Gemini, or ChatGPT. Describe the task to it the way you would explain it to a new staff member or a study partner on day one. Do not ask for code. Just have the conversation. See what comes back.

That is the entire on-ramp. The 5 hours of manual labour is where my version started. Yours is somewhere on your calendar this week.

If you have a ‘5-hour task’ but don’t have the 2 days to talk to Gemini, that’s what we do at Kade Labs. We build the bridge so you can get your morning back. Get in touch.