Writing
Notes from the build. Usually about something that took too long to figure out.
Rust Is the AI Era's Language
AI removes the implementation bottleneck. What remains is verification, and Rust's compiler has always been built for that.
Read →Don't Let AI Write Your Docs - Let It Interview Your Team
AI-generated context files hurt more than they help. The fix is using arc42 as a structure and the LLM as an interviewer.
Read →Engineering Quality as an AI Skill
A Claude Code skill that walks a structured engineering quality checklist, generates a status report, and tracks gaps iteratively.
Read →Dev Tools Are Going Headless
IDE plugins are being replaced by AI skills. The tooling is following the agent.
Read →AI Agents 101: A Glossary
Plain definitions for the terms you keep seeing - agent, loop, memory, skills - without the hype.
Read →Your financial data model is a contract, not a convenience
Schema decisions that feel small when building a PFM or investment analytics product - and become rebuilds when you get them wrong.
Read →Cheap verification is your biggest unlock when building ML products
The reason AI works where it works and fails where it fails maps almost exactly onto whether the domain has cheap verification. This is the hidden variable - not model size, not architecture.
Read →Parser generators: the unglamorous tool that beats LLMs at structured parsing
For constrained-format data like SMS alerts, regex breaks down and LLMs are overkill. Parser generators give you 40x the throughput, exact outputs, and logic you can actually debug.
Read →Choose Your MLOps Platform by Team Bottleneck, Not Feature Matrix
The right MLOps platform question isn't which platform is best - it's which team function is your actual bottleneck.
Read →Stop Saving CSVs to S3
CSV is a clipboard, not a storage format. Here's why data science teams should switch to Parquet - and how row groups make large-file processing on S3 actually work.
Read →