TL;DR: I am building Esuite with AI agents not as your typical software engineer, but with the ability to see the machine behind the mess, and the discipline to direct agents precisely.
The first thing I ever built in Python was a ping-pong game. The ball hit the roof and kept going. It never bounced back. So I also bounced off the course.
Six and a half months ago I made my first commit on the Esuite repo, the AI business consultant I'm building for Africa's solo, micro and small businesses. Today it runs on database migrations, API routes, scheduled jobs and 686 passing tests. It also keeps a public changelog that tells on me every time I change my mind.
This is the technical story of a Bachelor of Agriculture degree holder turned designer building production software with AI agents. It's also about why I threw away working code more than once, first unknowingly, subsequent times on purpose.
The honest bit first
I am not a software engineer. I've started Python three times. The first time gave us the ball that went to space. The third time was Henley Business School's Python for Coding and Business Analytics, and this time the ball stayed. Add a stack of product management courses and the 25+ courses on my CV, and you get someone who can read a system but would rather not hand-write one.
See behind the process:
2017, a bread and events company. I built its accounting in Excel. That cleared the company's debts and the staff arrears. It also put me in hospital from overwork, but the spreadsheet balanced.
ADCOM, Abia State Diaspora Commission. I ran a scholarship funnel with gates, scoring and an audit trail that screened from 211 applicants to 40, then 10, then 5 awardees.
9Mileon. I redesigned a cooperative into a fintech. I first demonstrated the trust indices and credit scoring with live humans who had engaged me as their brand and strategy consultant, before adding tech to the guarantor flow and safe disbursement. A credit score is a state machine with consequences; get the transitions wrong and someone's lifeline disappears.
So when AI coding agents got good enough, I didn't need to become an engineer. I needed to become a very precise master over the agents.
How I actually build
I treat my AI agent like a senior cofounder who happens to type code super fast. I write compound briefs of five to twelve items: what the user sees, what the data must guarantee, what must never happen. Then I review the architecture it proposes before a line gets written. I read the agents thinking as it works, and sometimes stops or corrects it midway.
I debug questions, not code. When something doesn't look right, the question I am able to ask is which assumptions, not which line. So, I designed a changelog as part of my discipline; to see my thoughts flow over time.
Era 0: before the repo existed (2023 to early 2026)
Esuite started with my mother's soybean powder. She had the energy, ambition, good product, but it wasn't shelf-worthy, and it lost a nutrition deal because of that. I was barely a teen, and nobody could tell her what to fix.
At the earliest idea I thought Esuite was an ERP. It isn't. An ERP records what you did, whilst a consultant tells you what to do next and checks whether you did it. That reframe took two years of research in Lagos and Aba, multiple documents and design files and conversations with several business owners; few I intentionally reached out to, the majority I met naturally in my too many MSME events. Some of that early work sits in the changelog as "built before the repository existed".