// AI Engineering

Mature AI engineering does not package a model as a universal assistant. It places it inside a system with clear boundaries and risk control.

The real starting point is not model size or compute, but whether internal data is searchable, quotable, traceable, and answerable.

AI coding tools accelerate generation, but they also inflate review pressure. The scarce resource is no longer raw output, but the team’s ability to understand and validate that output.

Two mechanisms behind Context Rot — attention dilution and context poisoning — and the different answers the industry is giving. Why is "starting fresh" still the most effective fix?

Tokens are becoming the raw material of this era. Workers are like factories — buying raw materials, processing them, and adding value. Understanding this logic is key to finding your core competitive advantage.

Exploring WebGPU's potential for running deep learning models in the browser — from API design to shader optimization, building a real-time inference engine.