Posts
A chronological view of the writing here, newest first, older pieces still visible.
Elixir 1.20 brings gradual type checking into a language known for dynamic code, concurrency, and a light touch. That is not a defeat for dynamic languages. It is what mature engineering systems do when they need earlier feedback.
The GitHub token stealing and Red Hat npm compromise incidents are not just stories about bugs. They show why AI-era security needs evidence, verification, remediation, and prevention that people and agents can actually follow.
The terminal is not coming back because engineers are nostalgic. It is coming back because AI agents need an executable, observable, replayable workspace.
The practical use of AI in software testing is not to let a model decide what passed. It is to connect requirements, test cases, scripts, evidence, defects, and regression scope into one feedback loop.
AI makes one-off output cheap. The harder question is whether a company can turn those outputs into reusable judgment, better workflows, and institutional memory.
Agent security is not about making the system prompt sound stricter. It is about redesigning work authorization: outside content can be evidence, but it cannot grant permission to act.
DBOS argues a single Postgres replaces Temporal, Restate, and Step Functions for durable execution. Reading the mechanism of each system shows where that claim is exactly right and where it quietly isn't.
Sixty-one years separate Gordon Moore's 1965 paper and He Tingbo's 2026 paper on chip scaling. A careful reading of both, with the math each one actually proposed.