Latest writing
MIT’s AI Report: Protect the Struggle, Not the Grade
MIT’s AI committee says AI broke the proxies we use to measure learning. It rejects detectors and one campus-wide rule, and asks for AI-aware courses and assessment.
Read the post →More posts
-
Jev: An AI Model That Returns Types, Not Text
TypeSafe AI’s Jev is a “System One model”: it takes state and typed questions and returns calibrated probabilities instead of text, in one parallel pass. The speed claims are self-tested, but its evals hide a better...
-
Torvalds on AI: It Is a Tool, and 100% of Your Code Was Always Written by Compilers
Linus Torvalds says he gets angry hearing “99% of our code is AI”, because 100% of it was already written by compilers and nobody brags about that. AI is a productivity tool in the compiler lineage, and its real cost...
-
First Principles: LLM Agents (Most of Yours Are Chatbots Wearing a Hat)
LLM agent design is buried under framework worship, multi-agent fantasies, and vector-DB orthodoxy. Strip it back to what is provably true and only six things survive. Most of the architecture teams reach for is a...
-
So You Want To Build An LLM Agent
Most agent demos are toys: the tweet-thread prototype works, production breaks in week one. A vendor-neutral 30-line ReAct loop, the six failure modes that kill prototypes, and the patterns that actually save you...
-
Dependency cooldowns considered harmful: a first-principles analysis
Dependency cooldowns promise supply-chain safety by hiding releases for N days, but the math shows they guarantee extended CVE exposure while catching only a narrow band of attacks that lockfiles and staging already...
-
First Principles: Distributed Systems (Most of Yours Should Not Exist)
Distributed systems thinking is drowning in inherited wisdom — microservices by default, horizontal scaling as a reflex, eventual consistency as a catchphrase. Strip it down to six irreducible truths and most of that...
-
First Principles: Data Engineering and ETLs
ETL is not a law of nature, it is a convention born from constraints that no longer exist. Strip data engineering down to its irreducible truths and rebuild it from scratch. What emerges is a discipline centered on...
-
First Principles: DevOps
Strip away the Kubernetes hype, the “DevOps Engineer” job titles, and the cargo-culted CI/CD pipelines. What’s actually true about DevOps when you reason from first principles? Five irreducible truths, six design...
-
First Principles: Software Design
Strip away the design patterns, the SOLID principles, and the architecture astronautics. What’s actually true about software design when you reason from fundamentals? Five irreducible truths, three reconstruction...