Something I learned over the years: The implementation matters, only when you get the API design right in the first place.

Anyone checking this? Valdi from Snapchat. a cross-platform UI framework that delivers native performance in declarative TypeScript, and it compiles directly to native views on iOS, Android, and macOS—no web views, no JavaScript bridges. React like. https://t.co/cUmNVB4wbR

I like the idea, I am not fond of coupling this abstraction with a compiler…it is usually a lib (Effect.ts) or a service (Temporal). Convention over configuration in this case makes things hard to reasoning and debug. But I love how the Vercel team have this sort of exploring.

I have very different experiences here. I was having similar ideas until I have used Gemini 2.5 Pro extensively. It’s 1M context solves tons of problems, I have back and forth asking it to write code for a 100 lines prompt, and update instructions through chat. It did it every

@github But here's a potential blocker: The tool generates data models in paths like spec/001-develop-to-do-app/. That folder name is an illegal module name in Python, causing major downstream issues during implementation. 7/7

@github Heads up on the downsides: Scope creep is a real risk, and you can easily end up with massive design docs. This also has a direct impact on cost—since each layer has its own output, an expanding scope will burn through API tokens incredibly fast. 6/7

@github This is a lifesaver for new projects with tangled features. It helps you decompose the work and manage scope effectively, guiding you toward a future-proof architecture without over-engineering. 5/7

@github The layered approach promotes focus. Since everything is in Markdown, edits are simple. A tasks . md file automatically tracks the implementation journey, making it easy to see your progress every time you come back. 4/7

@github The process is methodical: features are broken down into layers (e.g., specify, clarify, plan, audit, implement). Each layer builds upon the last, adding depth and detail progressively. 3/7

@github This tool hits the sweet spot between a 'YOLO' development vibe and overly detailed prompt engineering. It leverages GenAI to infer and enrich your product design, which is a massive time-saver. 2/7

I am using https://t.co/9E75FHo5j3 from @github quite a lot recently. Share the experiences here: 1/7

I haven’t touch react router for quite a while since the documentation is a bit hard to follow among the 3 modes. And I am afraid the remix v3 situation puts the future of react router in a really uncertain state. So moving away from it seems to be a good choice.

For building SPA these days, which framework would you use? A decent SPA work needs to support route splitting, route level fetching for preventing waterfall and SSG support for before-auth part. So far, react router v7, tanstack-router and next.js all have this feature.

I must be using a different Sonnet 4.5. 1. It finished a task with a file containing type error. (Previously, it always run type check to make sure everything works, no additional prompts needed) 2. After refactoring, it doesn’t change all places…so the app crashed (never

I believe, the real hurdle between you and a fine-tuned model for your specific workflow is the time and energy to construct a quality data set. 🥹 Since it is more like you need to know the answers before asking questions than to “teach” the model.

any modern http app supports streaming? I mean, to show me chunk by chunk, instead of the final response, so far, only postman can do it.

Two feature every modern apps need: 1. Passkey 2. If no 1, then tell me what is the last provider I used to login, Google? Apple? GitHub? Microsoft? Twitter? Seriously, this is a big headache.

Wow, the latest `google/gemini-2.5-flash-lite-preview-09-2025` is just super fast, even faster than `x-ai/grok-4-fast:free`. I tested loads of times, and the time for streaming down the 1st token is within 800ms on OpenRouter. While `grok-4-fast:free` is like 2sec.

Have to say this is what I have experienced, the current Siri typing chat is actually really solid and better. But the long time muscle memory of opening an app for doing something is just unbeatable… Not to mention you might have some other needs that requires booting up an app

And of course, it finished before my session ends 😆 I am still in the gym with 2 sets of lateral raise to finish.