Yo, people who develop for the Apple App Store! I have written a library and tool called XCUIPromoKit that helps your agents design and produce promo assets for your app store listing (search photos, header photos, and promo videos). github.com/RISCfuture/XCUIPr…

If you’re lamenting the eventual death of Maestral, the macOS Dropbox client, I’ve got a replacement for you: zephyrmac.app Zephyr is a Mac-assed Dropbox client with Finder integration and a slew of Mac-specific features. Available for download and on the App Store.

The future of software is hyper-personal. Whoever happens to pay for dinner, everyone will just use their individual vibe-coded Splitwise knockoff.

All of my iOS/macOS apps now have slick Claude-designed marketing sites: riscfuture.github.io/FART/ hsk-hanzi.github.io/iOS/ cta-helper.github.io/iOS/ riscfuture.github.io/SubTrac… sf50-told.github.io/iOS/ riscfuture.github.io/IP-Inbo…

If it hasn’t already been done, I am coining the term “trad-coding” to mean the opposite of vibe-coding.

In the future, all performance reviews will just be reading what Claude stores about you in user_memory.md.

Lol, Anthropic registered “claude.md” as a domain. Clever.

I’m sorry, but you’re just a regular prompt writer, and this company needs 10× prompt writers.

A company I know just got “approved” to use AI after a “security review.” Which is ridiculous — no security team would ever approve the use of AI. AI is a security nightmare. The only reason we all use it is because if we don’t, we’d be out-competed by everyone else who does.

These AI companies should probably stop calling it “Thinking Mode” or whatever, since in reality, it’s still an LLM, and no thinking is actually happening.

It’s so quaint, these old time travel shows where they say not to touch anything in the past because “the timeline might be affected,” like our current timeline is so precious. If I ever end up in the past, I am shooting every dinosaur and stomping on every rodent I can find.

How is it that the ENTIRE Claude Code system prompt fits into just 9.4k tokens, but your typical SaaS MCP (Sentry, GitHub, etc.) somehow uses 30k tokens?? Is it that Anthropic's engineers are just that good, or do MCP engineers not prioritize efficient prompt engineering?

I vibe-coded a menulet that helps you track your Claude Code token usage: github.com/RISCfuture/Claude…

Man, there is no better feeling than preflighting the plane for a sunset joyride, while your inbox fills up with emails of GitHub CI failures because Claude is rage-committing trying to fix your broken pipeline. That used to be me, my AI friend. That used to be me.

Due to the overproliferation of icons, people using screen readers may actually have a better browsing experience. They get “Remove attachments over 20MB” while we have to stare at an icon of a paperclip with an inset bomb and anvil, trying to figure out what that means.

Cursor is like the perfect example of how VCs WAY over-indexed on AI startups. It seemed like a slam-dunk unicorn, raising almost $1B on $10B … but now that the industry has matured, it’s obvious that there's no moat and they’re up against a completely saturated AI coding market

Who else here is on Team #bypasspermissionsalways?

All these CEOs announcing eng layoffs “because AI makes my engineers more productive” is highly sus. If your engineers are more productive now, why are you doing the same with less? Why aren’t you being more ambitious and doing more?

We all know the saying “reduce, reuse recycle”. It’s three ways to avoid generating extra waste, in priority order. I’ve invented a similar saying for writing fault-tolerant systems: “recover, resume, retry.” Also in priority order.

Does the whole “You are a helpful assistant who…” prompt preamble actually make a difference, or is it just cargo-culted industry wide from one engineer who wrote it one time at OpenAI?