Building an Agentic Software Factory on a Brownfield Fintech Codebase
How Leanpay rewired its engineering organisation around agents — on legacy systems, in two regulated markets. The full story, shared so you don't have to figure it out alone.
About This Event
Since May 2026, agentic tooling stopped being an experiment at Leanpay and became how we ship.
Every feature now flows through our software factory. Humans frame, review, and judge — agents implement. The factory isn't finished; we rewrite skills and improve context knowledge daily. But the direction is irreversible.
This didn't happen on a greenfield. It happened on a legacy fintech codebase, in two regulated markets, under strict supervision, GDPR obligations, and full audit trail requirements.
We are not hosting this meetup to impress you. We are hosting it because it took us significant time, painful iteration, and a lot of failed attempts to get here. We want to compress that learning curve for other engineering teams. What follows is the honest story — failures included.
Topics:
🔧 Harness Engineering The software factory doesn't start with agents — it starts with decomposition. We'll show how we broke down a legacy fintech codebase into bounded context pieces, sized and structured so that agents have enough scope to be useful and tight enough boundaries to produce deterministic outcomes. Too broad, the agent drifts. Too narrow, you lose leverage. Getting that decomposition right is the foundational engineering problem nobody writes about — and the one that unlocks everything else.
📄 Context Engineering That Works Agents are only as good as the context you give them. We'll open our actual CLAUDE.md files — domain rules, compliance constraints, codebase conventions, decision boundaries — and show what separates a context file that guides an agent from one that does nothing. We'll also cover how our product team uses AI in day-to-day product operations specifically to generate and maintain that context — closing the loop between what the business knows and what the agent understands.
🔄 Managing the Transition: From Coders to Factory Operators and Harness Engineers. The hardest part of building a software factory isn't the technology — it's the people. We'll share how we managed the shift inside our own development teams: what changes when your engineers stop writing code and start designing, orchestrating, and judging agent output. What skills transfer, what needs to be unlearned, and how to lead a team through a transition that has no established playbook. Honest account, including the resistance.
🔭 The Agentic Bank: Our Vision We close with the bigger picture. What does a fully agentic bank actually look like? How do credit decisioning, compliance workflows, and product delivery change when agents are permanent members of every squad? And the conclusion we keep returning to: good engineers are more critical now than before — because someone has to design the factory, write the contracts, and judge the output. That someone is you.
Why We're Sharing This
We spent the first half of 2026 figuring this out. A lot of it the hard way. We would rather the engineering community learns from our mistakes than repeat them. Everything we present — the approaches that worked, the ones that didn't, the frameworks we built — is real, in production, today.
Speakers
Leanpay business and technology leadership, together with our external partners from Miura.
Who Should Come
Software engineers, tech leads, engineering managers, and anyone working in or curious about fintech and banking. Whether you are deep in an AI transition or just starting to think about it — this is for you.
AI skeptics are especially welcome. Bring your hardest questions. That is the point.
Format: Four topics (Serbian language planned)· Live demos · Open Q&A · Drinks · Networking
Admission: Free · Seats are limited
Location: Belgrade