Blog
Notes on data science, value investing, and building things.
Trust, but Mutate: When AI Tests Lie to You
A blind reviewer can prove code is clean, but not that it works. Tests do that, and TDD is nearly free now that code is cheap. But when an AI writes the tests, the only honest way to know they are real is mutation testing.
The Blind Reviewer: Context Engineering for AI Agents
Gates decide what code is allowed out. Context decides what goes in, and more context is not better. This post is about feeding an agent a small, focused, well connected context, and then checking the result with a reviewer that knows nothing about what you asked for.
Garbage In, Garbage Out: Why Your Repository Is the Real Model
An AI agent does not obey your prompt. It obeys your codebase. If the base of the project is weak, the model copies that weakness at high speed. The fix is not a better prompt or a second AI reviewer. The fix is deterministic checks that cannot be argued with.
When to Skip the Vector Database: A RAG vs. File Agent Benchmark
A tweet claiming plain text files navigated by an agent beat RAG sent me to the lab. The results are more nuanced than either side admits.
The Death of the Code Monkey: Welcome to the Era of Vibe Coding and Agents
AI agents are automating away mechanical coding tasks, widening the gap between junior and senior developers. The real question is not whether to adopt these tools, but whether you understand what to build and why.
Follow the Journey
Want to follow along? Reach out on LinkedIn or GitHub.