Bumping a PHP monorepo to 8.5: the mechanics
Sixty-seven packages, one CI matrix, one PHPStan target. What it takes to actually move the floor.
Ai
PHP 8.5 restraint: features we did not adopt
An upgrade is also a decision about what not to use. Property hooks, the pipe operator, and an array_find pass we mostly rejected.
Spec Kitty mission lifecycle: a domain modeling pass through Giiken
What a full Spec Kitty mission actually looks like end to end: spec, plan, tasks, implement, review, merge.
The PHP 8.5 deprecation sweep: from 34 warnings to zero
Three deprecation categories. Twenty-nine call sites. A test corpus that surfaces them all.
What a real AI-assisted PR looks like
PR
Build an eval harness for 184 AI agent prompts with promptfoo
How to build an LLM-as-judge eval system that scores AI agent prompts on quality, identity, and safety.
Prompting isn't engineering, it's storytelling
The best prompts don’t give instructions. They set a scene the model can reason inside.
He spent two years letting cars drive him. Now he's back on horses.
A man spent two years letting cars drive him around. He’s back on horses. 865 engineers debated whether that means the cars are broken.
Managing LLM context in a real application
How Claudriel manages LLM context in production: conversation trimming, turn budgets, model fallback, prompt caching, and per-turn token telemetry.
Vibe coding isn't the problem. Your workflow is.
The critics of vibe coding aren’t wrong about the symptoms — they’re wrong about the cause.