shpak.dev

Yeah, that’s something to think about. And if we look a little further into the future, I still think customers won’t need all those middlemen (CMS, hosting, and so on). They’ll simply interact with their website and change its design and content using a mouse and their voice. All of this will be in the cloud, of course. Something like (Wix / Shopify) + AI. And I doubt that websites themselves will remain in the form they are in now. The way we get information is changing...

I think we're getting close to having open-source CMS tools that are on the level of Gutenberg or Notion. For me that'd be the last hurdle. There's Dhub and Basehub now, but both would just add another subscription. I've already got pipeline set up that deploys changes to my theme when I merge PRs on Github, and now I'm replacing plugins with in-theme functionality so all of that benefits from the same pipeline. There's definitely a part of me that wants to leave WP and start with a clean slate, but I think I can get most of the benefit just by doing this stuff. I realize the other day that I'm not yet tapping all of Cloudflare's potential, and I can likely move my actual WP install to a much smaller and cheaper server.

Rethinking my own WordPress site. Generating HTML is so easy now. Friction comes translating that HTML to block editor, but is it? If you start with empty WordPress and have AI build a custom theme and a custom block plugin... then the pipeline starts to get really good.

I totally get what you're mean. We're currently at the beginning of this major transition period (WP -> AI), where it's already clear that things are about to change drastically. But it's not entirely clear yet how to apply this. In some cases, moving away from WP is already working well right now, but otherwise, Wordpress is here to stay for the long haul. I just optimized a WP site with an HTML body size of ~700 KB (uncompressed), and I’d like that experience to stay in the past :)

Been exploring this, and as of yet I still haven't been able to find a solution that lets me write content in an editor of the same caliber as Gutenberg for free. I know I could use Astro or a zillion other SSG frameworks for the site, but I don't want to write content in Markdown, mainly because I like being able to just paste images from my clipboard. Markdown would mean typing or pasting actual file paths, as far as I can tell.

@AnimalGeoLife Who leaked my ex's video? :)

So cute! 🐸🥰

Men in the past: - Oh, look, that wolf tried to eat me, but I was stronger! Men in 2026: - *crying and tears* Claude Code has been glitching for days now - what should I do? Mom, help me!

Many people don’t realize how important it is to monitor server performance and analyze raw logs. This is the key to quickly improving your website’s search rankings. It’s the first step. You need to know: response times, HTTP statuses, slow requests, broken requests, and response sizes. You need to filter requests correctly and take into account Min, Max, p50, p95, Std Dev, and CV. This is serious but necessary work.

It seems the era of "flat web pages" and interfaces is finally coming to an end. We can expect total 3D immersion (like in video games) and plenty of augmented reality.

@stevibe Is there something like this in llmfit?

Which local models can actually handle tool calling? I built a framework to find out. 15 scenarios. 12 tools. Mocked responses. Temperature 0. No cherry-picking. Tested every Qwen3.5 size from 0.8B to 397B, and since some of you asked after the distillation tests: yes, I included Jackrong's Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled too. Only two models went all green: the 27B dense and the distilled 27B. The 397B? Failed two tests. The 122B? Failed one. The 35B? Failed two. The timed-out results — mostly on the smaller models, are cases where the model got stuck in a loop, repeating the same tool call until it hit the 30-second limit. The test that exposed the most models: "Search for Iceland's population, then calculate 2% of it." Simple, but 35B, 122B, and 397B all used a rounded number from memory instead of the actual search result. They didn't trust their own tool output. Small models hallucinate data. Big models ignore data. The 27B just threaded it through.

@jjack_arturo Many things in WordPress have remained stuck in the past, back in 2010. No wonder

It's annoying that #WordPress auto-installs a new theme every year, which then needs updates installed until you manually delete it. Isn't this potentially a threat vector?

@SNagarani1419 It's always been the same - to outrun a bear, you don't have to run faster than it, you just have to make sure you're not the last one :)

if ai replaces your current job, what you will do ?

@TTrimoreau This is the end

What's the most important thing you've learned from Al?

It’s so nice to see how Claude Cowork, using Chrome, debugged and found a solution in just 20 minutes for a bug on frontend that had been troubling the website for a couple of years! No one wanted to dive into megabytes of CSS and JS code to fix it. Claude did it while I was out having dinner.

Watch how fast Gemini 3.1 Flash-Lite can generate websites. ⚡ This browser creates each page in real-time as you click, search, and navigate. Give it a try → goo.gle/4t9In1R

@measure_plan A woman was supposed to walk into the room and ask: “Son, when are you going to get married?” :)

made a polyrhythm visualizer (tonejs, threejs, mediapipe)

@DavidOndrej1 You forgot to add “or I'll block you on OnlyFans” to the prompt :)

Claude Dispatch is actually AGI.

I really love that - thanks to AI, many useful apps, services, and tools are being rewritten and are getting faster - some by x2, others by x100. And that’s a free bonus for all of us. it’s exactly what we love. These really are exciting times!

@aarondfrancis just released Solo Term - It's such a clean and pleasant thing! I'm going to give it a try. I think this is what a product should look like when an good engineer creates something for another engineer and puts a much love into it 😍

@masonwarner Take off your shoes indoors? :)

M5 Max 128gb of RAM lfgggg What should I do first?

@weswinder I came here to point out that documentation actually exists. But someone was quicker than me :)

claude code hack you can run bash commands directly in claude code just press "!" then type whatever bash you want to run handy if you don't want to leave the claude code tui