noun · first recorded 2026 · status: emerging
Loopcraft
1.The practice of wrapping an AI agent in schedules, verifiers, subagents, and enough bash that it can keep prompting itself after you go to bed. Prompt engineering, but the prompt now has a night shift and a cloud bill.
At 11:47 p.m. in Hayes Valley, Maya called her cron-triggered Claude reviewer "loopcraft," closed the laptop, and woke up to 38 pull requests and one apology.
01Origin
The underlying discourse arrived before the name. In early June 2026, Claude Code lead Boris Cherny described moving from prompting Claude to writing loops that prompt it, Peter Steinberger told coding-agent users to design loops instead of prompts, and Addy Osmani published "Loop Engineering" on June 7. Osmani defined it as replacing yourself as the agent's prompter with schedules, state, isolated workspaces, tools, and verification. ([substack.com](https://substack.com/%40dewilliamsco/note/c-293289726?utm_source=openai))
On June 12, swyx packaged the idea as "Loopcraft: The Art of Stacking Loops." His framing was less about cron worship than abstraction: move down into tighter loops for reliability and up into larger loops for leverage. LangChain cited the phrase four days later while describing a stack of agent, verification, event-driven, and self-improvement loops, giving the coinage at least one respectable institutional witness before the timeline moved on to another noun. ([jinminlee.com](https://jinminlee.com/posts/ai-agent-design-five-layers?utm_source=openai))
02Notable sightings
03Spread the ism
Every entry ships as a card. Download it, post it, cite your sources.