Recipes
Practical playbooks for working across models, agents, harnesses, and the small moves that turn prompts into finished work.
GPT-5.6 changed the plan
One excellent model built the workshop. GPT-5.6 was good enough that the second bench could not stay empty.
read the recipe →two native sessions
Welcome to Atelier
We renamed the app, redesigned the website, and committed to a sharper direction. Here's why - and what comes next.
What broke at GitHub on April 23, explained without jargon
GitHub silently undid 2,092 pull requests of code last week. Most explanations assume you already know what a merge queue is. This one doesn't.
I used Opus 4.7 for 24 hours. Here's what I actually found.
A day with Claude Opus 4.7. What felt different, what broke, and the four things my old code was silently depending on.
Never lose your place: claude continue and claude resume
Two built-in flags that turn Claude Code sessions into persistent workstreams. The version of this workflow I wish someone had shown me on day one.
Review PRs like a staff engineer
A Claude Code skill that catches real bugs, not style nits - ranked by blast radius with confidence filtering so you only see what matters.