Type: url
Original: https://jonghoonpark.com/2026/03/29/agentic-engineering
Status: generated
Summary
A Korean write-up of a conference talk in which a self-described conservative developer recounts rebuilding and releasing the K-DEVCON community site with Claude Code. It traces the move from vibe coding to agentic engineering, argues against economizing on tokens on the grounds that a developer’s cognitive flow is the more expensive resource and that debugging should be handed back to the agent rather than taken over by the developer, and presents test code as the mechanism that makes AI-written code trustworthy — covering architecture rules, coverage thresholds and static analysis in the test pipeline, not only logic. It reports concrete uses including two-track routing that serves crawlers pre-generated static OG HTML so SEO works without SSR, a weekly loop running a third-party SEO skill and feeding its findings back to the agent, Lighthouse output handed to the agent as JSON, and templating repeated prompts as Skills. It closes on the last 20% AI does not fill: AI supplies answers but only a developer with direct experience chooses the direction, and AI carries no responsibility.