One of the coolest shifts in coding is the ability to ask 5 different AI sessions to write 5 different solutions, then run them all side-by-side as a full app to see proof of what actually performs best. This could never have happened before AI. If you are a dev and you are still guessing which methods work better or going by your own feeling or intuition, you are not prepared for the level of scrutiny of your code that is about to come.

People putting 4 GPUs in the computer so that they can ask the local agent to run ffmpeg correctly is so 2026. 🚀

Fun Fact: The FDA only allows clinical trials for drug development if the drug is targeting a disease. Since aging is not a disease, clinical research is virtually non-existent when compared to other barriers to healthy living.

Everything has its time. If we had AI in the age of MySpace, we would have broken the web.

@Egge what is the roadmap for npub.cash? I see good improvements with v2 and v3 in the codebase but looks like the production environment is still on v1. @calle for NUT-20, have you guys decided whether clients discover/verify quotes against npub.cash or the mint? My understanding is that the mint has no by-pubkey discovery by design, and v3's /api/v2/wallet/quotes gives me the quote refs — but it's pull-only and assumes the server is up and honest when I ask. And there doesn't seem to be a way to discover all npub.cash servers out there. It locks my implementation against a fixed list of operators. I guess npub.cash could just sign encrypted Nostr events to the receiving user that contains all quote references so that we can download them and check against the mint. How about pushing encrypted quotes to the receiving user via Nostr? When you broker a NUT-20 quote, you could fire an encrypted event carrying the quote ref (quoteId, mintUrl, amount) to the recipient. Then a client like Amethyst discovers funds straight from the user's relays and verifies each ref against the mint — no dependency on the API being reachable, and the server's worst case shrinks to withholding (detectable), not custody. Then we wouldn't need to hard code querying all npub.cash operators out there.

Also, I need a decent outlined cashu icon. 🫤

Do these new 3-state chip make any sense?