Back to Interests

AI Thoughts

Exploring the frontier of LLMs, agents, and generative art.

The Invisible Walls That Let AI Run Free

A Global Survey of Sandbox Technology and Regulation

From gVisor to Firecracker, from the EU AI Act to Singapore's GenA.I. Sandbox — how the race to contain autonomous AI agents is reshaping the infrastructure of intelligence itself.

LLMs Are Not Just Models

They're the New Operating System

Distilling Andrej Karpathy's framework — dual-file architecture, three-stage training, System 1/2 cognition, and the LLM-as-OS thesis — into a technical argument about where large language models are heading.

What Spec-Driven Development Gets Wrong

And How to Fix It

Spec-driven development fails for the same reason all documentation-first initiatives fail: maintenance falls entirely on humans. Augment Code's Intent changes that by making both humans and agents responsible for keeping the spec honest.