AI ships the first 80%.Engineering rigor ships the rest.
I’m Muamer Seljubac — CTO at Agrionelink and a software engineer with 9+ years behind systems that banks, e-commerce and smallholder farmers rely on. I use AI coding tools every day, and I hold what they write to the same bar as core banking: server-side auth, validated inputs, errors that fail loudly and dependencies that actually exist.
18M
daily users served
Last 20%
where systems hold or fail
9+
Years shipping software
Backend, full-stack, platform
18M
Daily users served
Rakuten Ichiba Item Page
5M+
Banking users supported
10 banks across 6 countries
CTO
AI-native engineering
Agrionelink · since 2026
Work that had to hold.
Production systems where the last 20% mattered — auth boundaries, data integrity, failure handling — long before AI could write the first 80%.
Engineering standards as CTO, Agrionelink
I own architecture, security and the engineering bar for agri-data and rural credit products — including eK Score — used by MFIs and banks. AI-assisted code goes through the same gate as hand-written code: auth enforced on the server, external input validated at runtime, no silent failures, no unvetted packages.
Scale
Field → lender
end-to-end platform
Role
CTO
Outcome
Bank-grade
standard for every merge
Ichiba Item Page, Rakuten
Backend engineering on one of Japan’s busiest product pages. Merchant-facing work (Shop Calendar, Topics List) plus a migration of 15 API features off upstream GraphQL onto REST — the kind of contract discipline that keeps 18 million daily users from noticing anything changed.
Scale
18M
daily users
Role
Backend
Outcome
¥10B
projected 3-year ROI
Core banking across six countries
Microservices for personal finance and mobile banking used by ten banks. Crediting, loans and the APIs that sit in front of money — where an auth gap or a swallowed error is somebody’s salary, not a ticket.
Scale
5M+
registered users
Role
Backend
Outcome
10 banks
in 6 countries
Expense platform at a Swedish scale-up
Java, Groovy and Angular on a product going through rapid growth. Three payroll and accounting integrations — where weak validation of someone else’s payload breaks real money flows — plus UI work that made internal testing faster.
Scale
30k+
customers
Role
Full-stack
Outcome
3
payroll / accounting integrations
Nine years of systems that had to hold.
Core banking for 5 million users, an e-commerce item page hit by 18 million people a day, and now CTO at Agrionelink — setting the engineering bar for a team that writes code with AI.
Leading technology for an agritech and rural finance company, with a distributed team that builds with AI coding tools every day.
- Own technology strategy and delivery end to end: architecture, engineering, security and the product roadmap.
- Set the guardrails that let AI-assisted development speed the team up without lowering the bar: server-side auth, strict typing, runtime validation, reviewed dependencies.
- Ship agri-data and credit products, including eK Score, to the reliability and explainability standard banks and MFIs expect.
I learned rigor where bugs cost money.
I started in fintech, writing backend services for core banking platforms used by millions of people across Southeast Europe — the kind of software where a bug is not an inconvenience, it is somebody's salary.
From there: a Swedish scale-up, US healthcare and transport systems, then two and a half years in Tokyo on Rakuten's Ichiba item page — 18 million people a day. That is where I learned what actually breaks at scale, and what it takes to keep it from breaking.
Today I'm CTO at Agrionelink, working from Bosnia with a team spread across several countries, building agri-data and rural credit products for banks and MFIs. We write a lot of that code with AI. It has made us faster — and it has made the gap between a convincing demo and a production system wider than it has ever been.
Generated code tends to fail in the same few places: auth checked in the UI but not on the server, raw SQL, N+1 queries, errors swallowed by empty catch blocks, type casts in place of validation, packages that are unmaintained or don't exist. None of it is new. It is the same list core banking taught me, arriving faster. I write about closing those gaps without giving up the speed.
If you lead engineers who ship with AI — or you are working out how to — I'm always happy to compare notes. The form at the bottom reaches me directly.
“AI gets you to 80% in hours. The last 20% decides whether you scale or collapse.”
How I think
Govern AI, don’t reject it
AI gets software to 80% in hours, and I use it every day. The job is deciding what is allowed into production — and making that decision automatic.
The last 20% is the job
Auth boundaries, input validation, query shape, edge cases. That is where systems scale or collapse, and exactly where generated code is weakest.
Fail loudly
An empty catch that returns 200 OK is worse than a crash. Errors get logged, surfaced and stopped at a clear boundary — never quietly swallowed.
Guardrails over good intentions
Never rely on an agent, or a person, to remember best practice. Encode it: rules files, strict compiler flags, linters and CI checks that say no.
Toolkit
Leadership
AI-native engineering
Backend
Frontend
Data & Infrastructure
Notes on engineering rigor in the age of AI.
Hardening AI-generated code, engineering leadership and what actually breaks in production — written from the CTO seat.
Nothing published yet — this is where new pieces will land.
Shipping with AI?
Let's compare notes.
Whether you're a CTO, a tech lead or an engineer working out how to put AI-generated code into production safely — or you'd like me on a podcast or panel — write to me directly. I read everything and reply myself.