A library built the slow way: every prompt tested on real work, signed by the person who uses it, and versioned when it changes. Plus the writing and the podcast that explain why they work.
Press ⌘K anywhere to search everything.
Every one tested on real work, attributed to a person, versioned so you can see what changed and why.
Browse the libraryEssays and field notes on prompt craft — what holds up across model versions and what was always theatre.
Read the writingLong conversations with people who folded AI into an actual job, taking apart the workflows they rely on.
Hear an episodeThe library
Picked because they hold up across model versions and because the people who wrote them still use them every week.
Builds the narrative spine before any slide exists, then writes headlines that carry the argument on their own — so the deck still works when someone flicks through it without you.
RonyWorks out what you actually want from it, what the other person is likely protecting, and the first two sentences — the part everyone rehearses badly.
DineshFor a contract, policy, or report where getting it wrong costs you. Separates what it obliges you to do from what it merely says, and flags the clauses worth a second read.
Lena FischerForces the model to separate what the trace proves from what it's inferring, rank hypotheses by testability, and give you the cheapest experiment that discriminates between them.
MeghnaCollections
Curated stacks for a specific hour of your week, in the order they're meant to be run.
Six prompts covering what a marketing team actually ships in a week — positioning, the landing page, the deck, the leave-behind one-pager, outreach, and social that isn't about nothing.
RonyFour prompts for understanding something properly rather than recognising it — for students, for anyone onboarding into a new field, and for explaining it once you do.
MeghnaThe four prompts for the part of job hunting that is actually work: making a résumé match one posting, writing a letter that couldn't be sent to anyone else, and walking in prepared.
RonyFrom the blog
The definitive engineering guide to context engineering for autonomous AI agents. Lost in the middle curves, context rot, KV cache math, hybrid retrieval with RRF, contextual chunking, tiered working memory, tool state isolation, attention budgeting, and deterministic verification across modern agent stacks.
Community
5 contributors so far. No upvotes, no leaderboards, no streaks — prompts arrive through the form, get run cold by a human reviewer, and land with a name and a version number attached.