Before You Build Custom Models
Teams should not rush to fine-tune unless they have the data, the strategy, and the metrics that justify it. The intelligence of the model is rarely the problem. Context is.
Posts
RSS2026
The Eval Flywheel
The teams that get real value out of evals are the ones where everyone, tech and non-tech, practices eval-based engineering. The flywheel ties it all together.
Staying With the Hard Part
In a research-heavy environment facing uncertain times, new ideas that challenge the norm require quiet persistence.
2025
An Autopsy of Agent Orchestration
After building dozens of agent systems, I've distilled orchestration down to 8 fundamental primitives that keep appearing.
Reflections after a year of vibe coding
A year of coding with AI assistants changed how I think about programming, creativity, and the craft of building software.
Last 10 Years of ML Workloads & Platforms
Every era of ML has the same fundamental promise: take messy reality, transform it into signals, learn from those signals, and ship decisions into products.
2024
AI Engineering is (still) Systems Engineering
Despite all the hype, AI engineering follows the same fundamentals: interfaces, reliability, observability, and cost.
Gradients and Roses
The Little Prince taught me that what's essential is invisible to the eye. Years later, I realized gradient descent works the same way.
Building in the AI Age: A Framework for Meaning and Evolution
When machines can build, what does it mean to be a builder? A framework for finding meaning in the age of AI.