I Tested 12 AI Models on the Same Video. The Results Were Wild.
Native video, frame-based, open-weight, proprietary. One hallucinated. Four failed. Here's the full breakdown.
AI AI Agents
Jiazhen Zhu
Most AI projects don't fail because of the model. They fail at the messy middle: data pipelines, production infrastructure, organizational alignment. I've spent 15+ years working in that middle, and I write about what I've learned.
Currently at Airbnb. Previously at Capital One and Walmart Global Tech. MBA from NYU Stern. MS in Analytics from Georgia Tech. Adjunct faculty at Northeastern University.
What I Write About
AI Productivity Systems
Building workflows that make AI agents genuinely useful in day-to-day work.
Data & ML Infrastructure
Lessons from building pipelines, platforms, and ML systems at scale.
Shipping AI to Production
What it actually takes to move from prototype to reliable, running systems.
Startup & Entrepreneurship
Notes on building products, finding ideas, and the lessons from YC and beyond.
Career & Mindset
Navigating tech careers, staying sharp, and thinking clearly under uncertainty.
AI Strategy for Leaders
How non-technical leaders can evaluate, adopt, and invest in AI with confidence.
Recent Posts
View all →Claude Mythos Leaked. Here's What a $450/Month Power User Is Actually Thinking About.
It's not "how good is it." It's "what can I attempt now that I couldn't before."
AI AI Productivity AI Agents
I Ran One AI Task Through 4 Harness Architectures. Here's What Broke.
I ran one AI task through 4 harness architectures — solo agent, generator+evaluator, and full planner pipeline. Here's what broke and what worked.
AI Productivity AI AI Agents Data/ML Infra Claude Code Developer Tools