Can an AI agent really replace staff work, and what did it cost or save? In one documented case, yes — a single AI agent absorbed the repetitive work previously handled by a five-person team. This is the account of Kenny Nwokoye, a Lagos-based business trainer and founder of Kenny Nwokoye LTD, who built his first AI agent on 29 January 2026 without writing a line of code, then let go of five roles whose day-to-day work was largely repetitive. He reports that his monthly team cost fell from about ₦1,050,000 to about ₦31,500 in AI tooling — a saving of roughly ₦1,018,500 per month, which he published alongside a Moniepoint business bank statement. This is one operator’s result in one specific business; it is a cost reduction, not an income promise, and results vary.
The situation: a small business carrying a lot of repetitive work
Kenny describes himself as explicitly non-technical: “I train people and build businesses — that’s my world.” He had used generative AI tools such as ChatGPT since 2022, but as a user of chat tools, not as someone building software. To keep his operation running, he employed a team covering five distinct roles:
- Personal assistant
- IT
- Community manager
- Social media manager
- Developer
Much of what these roles did day to day was repetitive busywork — the kind of recurring, rules-based tasks that fill a workweek without needing fresh human judgement each time. That distinction matters for understanding what happened next: it was the repetitive portion of the work that became a candidate for automation, not the entire scope of every role.
The intervention: building “Zuby,” an AI agent, with no code
In January 2026, Kenny discovered agentic AI — the idea that an AI system can be configured to carry out multi-step tasks on its own, rather than only responding turn-by-turn in a chat window. On 29 January 2026 he built his first AI agent, which he named Zuby, without writing a single line of code.
An AI agent, in general terms, is a system built on a language model that is given a goal, a set of instructions, and access to tools or actions, so it can complete work end to end rather than simply answering a question. What made Kenny’s build notable is that he approached it as a non-technical operator: he configured Zuby’s behaviour to take on tasks a person had previously done by hand.
What changed: which work the agent absorbed
According to Kenny’s account, Zuby took over the repetitive busywork that his team had previously handled across those five roles. Once the agent was absorbing that recurring work, he let go of the entire team in January 2026.
The framing Kenny uses is a shift in mental model — treating AI as infrastructure rather than as a tool you occasionally open. In practice, that means an agent runs assigned tasks continuously in the background instead of a person executing them manually. It is worth being precise about scope: the reported outcome describes repetitive tasks being absorbed by one configured agent in one business, not the wholesale, judgement-free elimination of every function a human team performs.
The measured result
Kenny published a before-and-after using a Moniepoint business bank statement. The figures he reported are:
| Item | Amount (per month) |
|---|---|
| Team cost before | about ₦1,050,000 |
| AI tooling cost after | about ₦31,500 |
| Reported saving | about ₦1,018,500 |
These numbers describe a change in ongoing monthly cost — what he was spending on staffing versus what he now spends on AI tooling to do the equivalent repetitive work.
Honest caveats
This case study should be read carefully, and with the following limits stated plainly:
- It is one operator’s result in one specific business. Kenny’s outcome reflects his own workload, his own team structure, and his own tasks. A different business with different work will not necessarily see the same result.
- It is a cost reduction, not an income guarantee. The reported figures describe money saved on staffing, not money earned. Nothing here promises revenue, profit, or any financial outcome to anyone who builds an agent.
- Results vary, and nothing is guaranteed. Whether an agent can absorb a given task depends heavily on how repetitive and rules-based that task is.
- It took building and iteration. This was not an instant switch; it involved building an agent and configuring its behaviour to do specific work reliably.
The honest takeaway is narrow but real: a non-technical operator built one AI agent that took over the repetitive portion of a five-person team’s work, and he reports a substantial reduction in his own monthly costs. It is a demonstration of what became possible for one person — not a template that guarantees the same numbers for anyone else.
Where this comes from
Kenny teaches the build approach behind this story through the AI Agents Mastery Masterclass, a live online programme for non-technical business owners and professionals who want to build and deploy their own agents. You can read more about the programme and how registration works at aiagentsmastery.me.