Insights
Notes from our own projects on AI agents, testing and small models. Each piece names its sources and the limits of what we measured.
September 24, 2026
Chihoon Shin · Insight · Small models
3 min read
Insight 02
A small, narrow judge that decides in about 2 ms can sit in front of a rule-based gate. We built one for a demo game and share what we saw and where it stops.
September 23, 2026
Chihoon Shin · The Last Mile of AI · AI agents
1 min read
The Last Mile of AI · 01
Small mistakes add up across a job. Our own engine was 90.7% right screen by screen but finished 8 of 12 tasks. Check where a mistake can no longer be undone.
September 23, 2026
Chihoon Shin · Insight · AI evaluation
5 min read
Insight 01
Two teams can use the same AI model and the same prompt and still get different results. We look at where that gap comes from and which failures are worth a person's time.
