commandrefresh

Welcome to commandrefresh

Benjamin — Tue, 07 Jul 2026 00:00:00 GMT

Every good workflow has a reflex baked into it. For a lot of us that reflex is ⌘R — the keystroke you hit without thinking when something feels stale and you want the freshest version of reality. That instinct is the whole idea behind this site. commandrefresh is a place to reload your view of the tech world, with a particular soft spot for everything Apple.

What you’ll find here

We’re starting narrow and going deep. The plan is a steady stream of writing across a handful of beats:

Apple — silicon, hardware launches, and the software that ties the ecosystem together.
Software & Hardware — the tools, machines, and peripherals worth your attention (and the ones that aren’t).
AI — practical, grounded coverage of where machine learning is actually showing up in the products you use.
Tutorials — short, buildable guides you can follow start to finish.
News & Opinion — the quick hits, plus the occasional argument.

No slideshows, no ten-page reviews that bury the verdict. The goal is signal.

Why now

The tech press has drifted toward two extremes: breathless launch-day regurgitation on one end, and endless hot-take churn on the other. There’s a gap in the middle for writing that’s fast but still considered — coverage that respects both your time and your intelligence. That’s the space we want to live in.

We’re also building this site in the open. It runs on a modern static stack (Astro, if you’re curious), it’s fast by default, and it’s a genuine pleasure to read on any device. Over the coming weeks you’ll see search, related posts, and a comments system land — but the foundation is here today.

The house style

A few promises. We’ll tell you when something is a guess. We’ll link out to primary sources instead of paraphrasing them badly. And when we get something wrong, we’ll say so and fix it in the open rather than quietly editing history.

If that sounds like your kind of tech blog, you’re in the right place. Bookmark the RSS feed, poke around the categories, and check back often.

Hit refresh. There’s always something new.

On-device AI is the quiet iPhone story

Benjamin — Thu, 02 Jul 2026 00:00:00 GMT

Most AI headlines are about models that live in someone else’s data center. The more consequential shift is happening in your pocket, where a growing share of the intelligence runs locally and never touches a network. It’s less flashy than a chatbot that writes sonnets, and it matters more.

Why local beats the cloud for the boring stuff

A huge amount of what makes a phone feel smart is small, constant, and personal: sorting notifications, surfacing the right photo, transcribing a voice memo, suggesting the next word. None of that benefits from a round trip to a server — it benefits from being instant and private.

Running inference on-device buys you three things at once:

Latency — no network hop, so results feel immediate.
Privacy — your data stays on hardware you control.
Availability — it works in airplane mode, on the subway, everywhere.

That combination is hard to beat, and it’s why the Neural Engine has quietly become one of the most important parts of the chip.

The privacy math actually changes

There’s a real architectural difference between “we promise not to look” and “it physically never left the device.” When a feature runs locally, the privacy guarantee isn’t a policy you have to trust — it’s a property of the system. For sensitive categories like health, messages, and photos, that distinction is the whole game.

The hybrid approaches are the pragmatic middle ground: handle the common cases on device, and escalate only the genuinely hard requests to a hardened server environment with verifiable guarantees. Done well, you rarely notice the handoff.

The constraint is also the design

Local models are small models, and small models force discipline. You can’t lean on a trillion parameters to paper over a vague feature, so the work shifts to figuring out exactly which narrow, well-defined problems are worth solving. That constraint tends to produce features that feel considered rather than bolted on.

What to watch

The interesting metric over the next couple of years won’t be benchmark scores. It’ll be how much capability vendors can push onto the device without wrecking battery life or thermals — and how honest they are about where the line between local and cloud actually sits. The companies that get specific about that will earn trust. The ones that stay vague won’t.

The best AI on your phone is the kind you never have to think about, running on hardware you already own, on data that never leaves your hand.

Apple Silicon's M-series: five years of quiet dominance

Benjamin — Tue, 30 Jun 2026 00:00:00 GMT

The M-series didn't just move Apple off Intel — it reset expectations for what a laptop chip should feel like. A look at where things stand.

Build a fast Astro blog in 20 minutes

Benjamin — Wed, 24 Jun 2026 00:00:00 GMT

Astro has quietly become the default answer for content-heavy sites, and once you’ve shipped one it’s easy to see why: you write mostly HTML and Markdown, you ship almost no JavaScript, and the thing is fast without you fighting for it. Here’s the shortest path from empty folder to a working blog.

1. Scaffold the project

Start from the minimal template so nothing is in your way:

npm create astro@latest my-blog -- --template minimal --typescript strict
cd my-blog

You get a tiny project: a config file, a src/pages directory, and not much else. That emptiness is the point — you add only what you need.

2. Define a content collection

This is the piece that makes Astro click. Instead of loose Markdown files, you describe the shape of your content with a Zod schema and get type safety plus validation for free:

// src/content.config.ts
import { glob } from 'astro/loaders';
import { defineCollection, z } from 'astro:content';

const posts = defineCollection({
  loader: glob({ pattern: '**/*.md', base: './src/content/posts' }),
  schema: z.object({
    title: z.string(),
    pubDate: z.coerce.date(),
    draft: z.boolean().default(false),
  }),
});

export const collections = { posts };

Now every post is validated at build time. Forget a title? The build fails loudly instead of shipping a broken page.

3. Render a post

A dynamic route pulls an entry and renders it. The render() call hands you a Content component that outputs the Markdown body:

---
import { getCollection, render } from 'astro:content';

export async function getStaticPaths() {
  const posts = await getCollection('posts');
  return posts.map((post) => ({ params: { slug: post.id }, props: { post } }));
}

const { post } = Astro.props;
const { Content } = await render(post);
---
<article>
  <h1>{post.data.title}</h1>
  <Content />
</article>

4. Build and ship

npm run build

You get a dist/ folder of static HTML you can drop on any host — no server, no runtime, no cold starts. The whole site is pre-rendered.

Where to go next

From here the additions are small and composable: an RSS feed via @astrojs/rss, a sitemap integration, Tailwind through the Vite plugin, and pagination using Astro’s built-in paginate() helper. Each one is a few lines, and none of them slow the site down. That restraint — adding capability without adding weight — is the whole reason Astro is worth learning.

Mechanical keyboards in 2026: a buyer's field guide

Benjamin — Fri, 19 Jun 2026 00:00:00 GMT

A few years ago, getting into mechanical keyboards meant group buys, soldering irons, and a lot of waiting. In 2026 the scene has grown up. Hot-swap boards are everywhere, quality is high even at the low end, and you can build something that feels genuinely custom in an afternoon. Here’s how to choose without drowning in options.

Start with the switch, not the board

The switch is what you actually feel, so decide this first. Everything else is downstream.

Linear — smooth top to bottom, no bump. Quiet, fast, and the default for people who type a lot or game.
Tactile — a bump partway down tells your finger the key registered. The best all-rounder for typists.
Clicky — the loud, nostalgic ones. Great feel, terrible office etiquette.

If you’re unsure, get a cheap switch tester before you commit. Ten minutes of pressing switches will teach you more than ten reviews.

Hot-swap is non-negotiable now

A hot-swap board lets you pull and replace switches by hand, no soldering. Given that the feature adds almost nothing to the price these days, there’s no good reason to buy a soldered board as your first one. It turns an irreversible decision into a five-minute experiment.

Layout: buy the size you’ll actually use

Smaller looks cooler in photos; bigger is often what you need in practice.

Full-size / TKL — everything, or everything minus the numpad. Safe.
75% — the current darling. Compact but keeps arrows and a function row.
60% and under — clean, but you’re learning key combos for arrows. Know that going in.

The 75% layout has become the sweet spot for a reason: it reclaims desk space without asking you to relearn how you type.

Where the money actually goes

You’ll hear a lot about keycaps and cases, but the single biggest upgrade in feel per dollar is a well-built board with decent foam and a gasket mount. That’s what produces the deep, muted “thock” everyone chases — not expensive keycaps. Spend there first, decorate later.

A sane first build

If you want a shortcut: a mid-range hot-swap 75% board, a set of pre-lubed tactile switches, and PBT keycaps in a colorway you like. That’s a setup you’ll be happy with for years, and every part of it can be swapped later as your taste sharpens. Start there, resist the urge to buy five boards at once, and enjoy the typing.