An AI Agent Is an LLM With a Harness
Jensen Huang's practical definition of an AI agent, why the harness matters more than most demos admit, and what this means for building agents at Good Journey.

🌍 USA | Singapore | Vietnam
We build software that moves people. If you've ever booked an airport transfer, there's a chance our team helped get you from A to B after your flight landed.
I'm the co-founder of GoodJourney.io — a ground transportation operating system. We handle bookings, drivers, vehicles, and a dispatch engine that matches the right ride to the right driver. Our focus is airport transfers: wherever you land, GoodJourney OS should get you where you need to go. We're working to unify ground transportation companies and online booking agents under one system.
We're a small team. We call ourselves team-silent — we keep our heads down, stay quiet, and get things done.
Right now I'm deep into learning how AI can automate my life, help me level up as a founder, coder, designer, and product manager, and how we can build the future of AI-powered applications.
When I'm not doing that, I practice my golf swing. I also eat too many Undercover Snacks dark chocolate quinoa sea salt mini crisps. They sound healthy. They are not.
This is my journal.
Jensen Huang's practical definition of an AI agent, why the harness matters more than most demos admit, and what this means for building agents at Good Journey.
Kimi K3 shows how quickly open-weight AI is catching the closed frontier and why that shift benefits developers, AI companies, and businesses like Good Journey.
A deep dive into Google's DESIGN.md format: what it is, why it matters, how it differs from AGENTS.md, and a hands-on tutorial building a landing page from a real production design system.
A practical guide to wiring Graphify into OpenCode so your AI coding workflow has persistent, queryable architectural context instead of session-by-session amnesia.
How Figma Make collapsed the gap between idea and usable interface, the economics of credits/tokens, and lessons learned from spending $15-20k worth of tokens during the experimental phase