Ca devrait être illégal de demander un appel téléphonique pour annuler un abonnement pour un SaaS @Payfit

Looking for modern iOS designs resources, ideally formatted in a way that make it easy for an agent to search/get/analyse. Any recommendations?

Research with llm is the fun part of building. Asked Fable to find some existing math/research to optimise token usage and it drew a parallel between quota usage optimisation and airline seats management. Also, it's using ski rental prediction models to improve ttl choice.

Stop pasting secrets in Claude Code, use mantaui.com > Secrets stored on your box in a file > Your agent knows how to access it safely > No transcript leak > No Claude Code telling 100 times to rotate it

Can't even debug your router config with Opus 5 🙄

Who's building an open-source version of this?

The best vibecoding trick to 10x output quality is to ask your agent to do some "online research for sota on this, best practices, best UX/UI", etc...

Just vibe-reversed-engineered a walking treadmill: I bought one, as a walking desk, that has an ios app + a physical remote. Starting with the remote is more convenient as I don't want to open up an app for that. But ofc, if I do that, the app doesn't recognise it, doesn't log it and doesn't sync the steps with apple health (which is all I want). In a couple of hours Claude: - figured out the protocol used to start & control the treadmill remotely, including quirks introduced by the maker of the treadmill - built a mac app that displays timer, steps, speed control, etc... on my status bar - built an ios app that syncs with the mac app to receive step data when I'm done (apparently healthkit is just available on mobile) and sends it to apple health I probably would have been able to do it by myself eventually but it would have taken me a week or so for probably a not as good result (though i would have had more fun).

Swift AI coding seems to be a lot slower and more of a hassle than just web, anyone else noticing that? Any tips? Or is everyone just doing react native?

Best tip to maximise token efficiency on a Claude Code/Codex subscription?

The combination of strong spec + strong static deterministic guardrails + weak & cheap models beats throwing blindly the strongest model to the problem. I've been using @MulticaAI to help with that and it's pretty amazing. Running it across a few different projects with a similar setup for each: 1 PM agent: the only one I'm interfacing directly with, it coordinates all other agents, split big issues into epics/sub-issues. Other agents can only hand off to it and it's the only one allowed to assign to me. It also has access to a browser to do e2e tests when needed. Running Opus 4.8 Multiple implementer agents: specialised agents that work on specific codebase part, some projects have only one, some others have more (frontend/backend/ai-pipeline for example). Running Sonnnet 5 1 reviewer agent: reviews PRs, back and forth with implementers, then hands off to PM. Running Opus 4.8 They bounce tasks between each others, commenting each ticket, leaving perfect context for the next agent. I'm mostly working from the inbox view, watching as everything is being worked on. Then from my Claude session, I spec things, create issues, assign to pm, and decide what/when to deploy to prod. No implementation at all here.