Taeyeon Kwon 권태연

platform engineering lead · seoul

I lead platform engineering at a multi-strategy asset manager in Seoul, where my team runs its delivery on coding agents inside a regulated financial environment. Most of my work is the machinery that makes that safe and honest: evaluation gates, approval surfaces designed to teach, and the event infrastructure underneath.

I write here about the operating model behind enterprise AI adoption — what it actually takes for a team to ship with agents and still understand what it owns.

Writing

The real cost of vibe coding is cognitive debt
Why I harden approval gates instead of removing them — and design them to teach the system they guard.
essay · august 2026

Work

gefjon-growth
A production hiring-automation system: a config-driven, multi-agent pipeline from intake to interview kits. Its per-stage workflow engine was committed in August 2025, two months before Agent Skills launched — the history preserves the dates.
public mirror · python · agentic workflows
banga-engine
방아 — a two-tier event-queue engine: a crash-safe, fsync-first JSONL journal with per-class coalescing and a peek-only screener census. 386 tests, all offline.
event infrastructure · python · 386 tests
yimir-root
The July-2025 template the hiring system grew from: an AI-native knowledge-base and workflow skeleton from the Gemini-CLI era, kept as a dated record of where the approach started.
workflow template · july 2025