I started in engineering, crossed into product, and ended up spending most of my time on agentic systems and the evals that keep them honest. That's the loudest part of my work right now, but not the only part — I've shipped across checkout, identity, and commerce at scale, and I care a lot about the unsexy stuff like measurement, trust, and what “working” actually means in production. Outside of work, I run half-marathons (slowly, happily), read more than is probably reasonable, and have a well-documented soft spot for elephants.

2019
2020
2021
2024
2025
2026
2025
2024

Notes and essays, filed by subject. Click a folder to open it.

poloclub.github.io

Transformer Explainer

The goal is not to build AI that does what we say, but AI that does what we mean.

lilianweng.github.io

LLM Powered Autonomous Agents

blog.pragmaticengineer.com

The Product-Minded Engineer

A product is a promise of value to be delivered, a commitment to solve a real problem.

svpg.com

Product Discovery

frankchimero.com

The Shape of Design

Design is not just what it looks like and feels like. Design is how it works.

alistapart.com

Dao of Web Design

paulgraham.com

How to Do Great Work

The unexamined life is not worth living.

fs.blog

The Map is Not the Territory

Chip Huyen on the gap between demos and production. The most grounded take on evaluation I've found.

Situational Awarenesssituational-awareness.aiMay 2026

Long, opinionated, and genuinely useful for thinking about where AI development is headed at the macro level.

The Product-Minded Engineerblog.pragmaticengineer.comApr 2026

Describes the exact mindset I try to hire for. Good framing for PM-eng collaboration too.

How to Do Great Workpaulgraham.comMar 2026

PG at his most earnest. The advice about following your curiosity over prestige is the one I keep coming back to.

Agentslilian.github.ioFeb 2026

Lilian Weng's definitive survey of LLM agent architectures. Dense, precise, worth re-reading every 6 months.

The Bitter Lessonincompleteideas.netJan 2026

Sutton's argument that general methods leveraging compute always win. Required reading before any architecture debate.