6 min read
The model is not the product
Everyone debates which model is smartest. Almost nobody talks about the ninety other things standing between a prompt and a shipped feature.
← All posts
AI19 posts
AI
Working notes on building with AI — models as tools, not magic: workflows, agent ceilings, evaluation, and the unglamorous systems that actually ship.
Written by Nzubechukwu Cyprian — Pharmacy student at UNN, full-stack developer and AI builder.
5 min read
My AI workflow is boring, and that's the point
No autonomous agents rewriting my codebase. No prompt wizards. Just a disciplined loop between a draft, a review, and a test suite.
6 min read
Prompting is requirements gathering with worse vocabulary
The teams that get good results from AI aren't better at talking to models. They're better at saying what they want.
7 min read
I gave an AI agent shell access. Here's what I learned.
A week of running an autonomous coding agent on a real project taught me where the ceiling is — and it isn't intelligence.
7 min read
RAG is a data-cleaning problem wearing a trench coat
Most retrieval-augmented systems fail because of chunking, duplication, and stale documents — not because of embeddings.
6 min read
What AI can't do for a solo developer
AI writes code fast. It still can't decide what's worth building, notice the bug you haven't hit yet, or care about your users.
5 min read
Evaluating AI output like a code reviewer, not a fan
A practical checklist for reviewing machine-written code: correctness first, then edge cases, then the quiet sins.
6 min read
Building with AI on a slow connection
AI-assisted development from Nigeria means timeouts, retries, and offline-first habits. Here's the setup that survives bad networks.
6 min read
AI is changing how beginners learn to code — badly
The path from confusion to competence used to run through being stuck. AI removes the being stuck, and with it the learning.
5 min read
The quiet rise of small, boring, useful AI
The most valuable AI systems I use aren't impressive. They classify, extract, and route — and they save hours a week.
6 min read
AI-generated code has a maintenance problem nobody prices in
Generating code got cheap. Owning it didn't. Every line still has to be understood by someone, eventually, at 2am.
6 min read
Testing AI features is different — here's my setup
You can't assert on a model's exact output. You can assert on structure, invariants, and failure behaviour. That's where tests go.
5 min read
Choosing models: a decision tree for people with budgets
The flagship model is rarely the right default. A practical way to pick: capability floor, then cost, then latency.
5 min read
Context windows are a design constraint, not a spec number
Bigger windows changed what's possible — and quietly changed how systems should be designed. Retrieved beats stuffed.
6 min read
What building AI products in Nigeria actually looks like
Payment rails, power, and latency shape every AI product decision here. Constraints don't stop building — they shape what gets built.
4 min read
Why I keep a prompt journal
A plain file of every prompt that worked, every one that failed, and why. The cheapest AI skill you can build.
6 min read
AI hallucination: a field guide from someone who ships
Hallucinations aren't random. They cluster in predictable places: numbers, citations, specifics, and anything you'd verify twice.
5 min read
The best AI feature is the one users don't notice
Generation demos impress. Inference that quietly fills gaps — autocomplete, defaults, cleanup — is what users keep.
7 min read
From notebook to production: the gap nobody blogs about
The demo works. Production is a different country. Version drift, cold starts, monitoring, and the user who types nothing you expected.