Learning from Every's Compound Engineering
Every's compound engineering breaks down AI coding into 4 patterns: planning (decoupling research from implementation) and executing work.
Drew DeVault draws a provocative parallel between Test-Driven Development cults and GenAI adoption, arguing both exploit developers' psychological need to feel competent while potentially undermining actual code quality.
Every's compound engineering breaks down AI coding into 4 patterns: planning (decoupling research from implementation) and executing work.
Unable to summarize - no content provided in the blog post.
Operations work is valuable and deserves respect, not disdain. Those who dismiss ops will face the consequences of poor operational outcomes.
Author discovered people find their content through LLMs, so they used AI tools to optimize for even better AI-driven discovery.
FLUX.2-klein-4B implemented in pure C. No additional details provided.
Files and open formats enable better data portability and longevity than siloed apps.
Made Code Wave for GitHub Game Off 2025 and shared the development process.
AI coding agents are causing quality degradation in code contributions, with users seemingly unaware of the problem they're creating.
Author added a WebAssembly compiler to Bob, their 15-year-old Scheme implementation project, which already includes Python and C++ VMs.
The author has a deeper appreciation for Mary Cassatt's art after becoming a parent themselves.
Cryptography has many pitfalls that catch developers off-guard when security algorithms contain unexpected vulnerabilities.
Don't rely on tools to do all your thinking—keep your mind sharp by thinking for yourself.
Parent received unexpected letter for toddler due to data breach, highlighting need for encryption to protect personal information online.
Podcast discussing Mac, Jobs, and HIG achievements with Ilya Birman, bidding farewell to Alan Dye, also mentioning Windows positively.
Using Claude Code slash commands to automate routine "quick requests" that would otherwise consume significant time and mental energy.
Developer avoided generating 1.5M lines of Go code by finding a better way to implement Unicode character properties in regexp for Quamina library.
Rust offers safer cryptography as 37% of crypto library vulnerabilities stem from memory safety issues rather than cryptographic flaws (27%).
AI agent ported MiniJinja (Rust template engine) to Go in ~10 hours with minimal human guidance, requiring only 45 min of active work.
A truly omniscient deity would prove itself through computational complexity, not faith—providing mathematically verifiable signatures rather than miracles.
Teams need both adaptation and optimization, not one or the other. Balance "explore" and "exploit" modes based on uncertainty and risk.