What AI Agent Benchmarks Actually Measure (And What They Don’t)
The agent that scored 72% on WebArena crashed repeatedly when a real user asked it to book a flight with a non-standard loyalty number in
September 30, 2026
My story began in Seoul, Korea, and my roots took hold across the Pacific, shaped by a childhood spent between cultures and a lifelong curiosity about how different worlds connect. My academic path eventually led me to the University of Michigan in Ann Arbor, where that curiosity only deepened.
For the last decade, I’ve channeled that curiosity into Web Design and SEO. I see websites as digital front doors — and my job is to make sure they’re welcoming, easy to find, and built to last. I love the moment when a business owner sees their vision finally come to life online, and I take pride in building the kind of technical foundation that can grow with them.
This blog is where I explore the next frontier: A.I. implementation and agentic workflows. We’re living through a fundamental shift in how work gets done, and I’m fascinated by how intelligent agents can handle the heavy lifting — freeing us up for the creative, human-centered work we actually enjoy. My goal here is to demystify AI and share practical strategies for building workflows that think, learn, and grow with you. Consider me your guide through this new landscape: someone who’s been on the ground floor of enough transitions to know that the best technology never loses its human touch.
The agent that scored 72% on WebArena crashed repeatedly when a real user asked it to book a flight with a non-standard loyalty number in
A team I spoke with last month was convinced their agentic customer support system was underperforming because their embedding model wasn’t good enough. They switched
What Anthropic Actually Changed Claude has always had guardrails around tool use, but recent updates to the model’s behavior — particularly in agentic contexts involving