You really don't need Next.js and its bloat, really you don't. svelte.dev/docs/svelte/... Or you know what else you can do if you're one of the "but Svelte isn't ready for production" people? developer.mozilla.org/en-US/docs/W... I've been talking to a wall about React/Next.js for 8 years 😂.

svelte/reactivity/window • Svelte Docs

svelte/reactivity/window • Svelte documentation

svelte.dev

Next.js@nextjs.org · 2d ago

When the network drops, requests normally throw. The experimental 𝚎𝚡𝚙𝚎𝚛𝚒𝚖𝚎𝚗𝚝𝚊𝚕.𝚞𝚜𝚎𝙾𝚏𝚏𝚕𝚒𝚗𝚎 flag keeps them pending and retries on reconnect. A companion 𝚞𝚜𝚎𝙾𝚏𝚏𝚕𝚒𝚗𝚎 hook reports when you're offline, so you can render some UI to let your users know.

LLM, "No speech in this video" I watch the video and hear speech. Me, "Hey LLM, I hear speech. Figure it out and extract the audio" LLM, "RMS: L −23.0 dBFS, R −23.0 dBFS → mono downmix −50.1 dBFS (27 dB collapse)" System, "We're good, audio extracts now" Me, ".... ok ..."

By far the biggest challenge in consolidating government data is different spellings for the same entity across multiple documents. Among the misspellings, the most common one is double spaces 😂.

I blocked bot access to a web app I'm working on, then pasted a link to a page into claude and asked it to review the JSOND snippets. Claude: `curl ...` Response: 403 Forbidden Claude `curl -A <insert-browser-UA>` Response: 200 OK 🤷

Come on Google, you can do it 😂! The reason for the not indexed pages is "Discovered – currently not indexed"... So close, yet so far.

AI bots have arrived... Facebook AI training, Semrush, and others. Cloudflare's settings to block them has exactly 0 impact. Nothing a WAF rule can't handle.

When you operate at tens of thousands of PDFs, even the tiniest improvement has a big impact. Tonight's fun: OCR memory consumption improvement

I read up on atproto.com/specs/lexicon, seems neat.

Lexicon - AT Protocol

A schema definition language.

atproto.com

Lunch time well spent to reduce my @railway.com memory spend 🤑. This is a job that runs whisper.cpp on hours-long meetings with many attendees. Changes: • Chunk a file and send it to whisper in windows • Clean up files loaded onto disk

Moonshot stole from Anthropic. Anthropic stole from everyone. So... What now 😂?!

If you ever need to transcribe 100s of 2h+ (up to 8h) videos with 7 to ~50 speakers and accurately apply speaker names and timestamps, AMA 😅. If you find a way to avoid this kind of work, avoid it 😂.

Oh my, OpenAI's desperation starts to get annoying 😂. Anthropic's too for that matter, what a clown show. Long live local, open-weight models.

"It's unacceptable that Chinese open source AI models may have been built on IP theft". They are so close. Now do OpenAI and Anthropic next.

75,884 pages indexed, without a doubt my biggest @svelte.dev web app. It's also the one that has generated the most revenue in the first four weeks since its soft-launch. I'm working with journalists to announce it. It's a slow but steadily moving process.

TFW when you finish a v1 -> v2 architecture rewrite, migrate 25k PDFs, 15M database rows, start the service and everything works. Immediate first thought: I'm looking at v1, this can't be right 😂. Second thought: Thorough test coverage is awesome! Claude worked for 35 hours on the migration ⏳.

Easy, upload the screenshot to AI and tell it to fix it. Ever since I vibe-coded seaquel.app, I learned AI is nothing more than a fast typewriter & rubber duck.

Seaquel - Database Management, Reimagined

Lightning-fast, resource-efficient database client with AI-powered assistance. Work offline, query smarter, and manage your data with unprecedented speed.

seaquel.app

Gergely Orosz@gergely.pragmaticengineer.com · 3w ago

All of this comes after Spotify was bragging how they use AI for software development, publish a massive number of PRs to prod per day. But reliablity has been going down for me. To the point that today this is what my portal looked:

"I broke one of your rules. I ran git checkout -- to compare test counts. Your instructions say no modifying git operations. It only reverted files I'd changed myself, and I re-applied the work — nothing of yours was lost — but I shouldn't have run it." CLAUDE.md instructions: meh, why bother?

Wow that's a first... Someone asked to have an early morning chat (7am) at a local coffee shop. I'm always up for early morning activities, so we met. His glasses felt off and when we ordered, I opened an app that scans for nearby Meta Glasses. Sure enough, it found one. 👇

"Scan the design docs in @docs/plans/ and list features that were deferred and haven't been implemented, categorised by impact and effort" You're welcome 🤝

One thing worth pointing out because I think it's incredibly awesome: The entire web UI is embedded in the Rust binary as a single HTML file, made possible by @svelte.dev 💪!

Mike@mikenikles.com · 4w ago

Introducing Orchestrator ⚡. A single-binary workflow engine that just works™. To scale horizontally, start the binary in worker mode on as many machines as you want. Orchestrator distributes the load automatically. github.com/webstonehq/o... 👇 Orchestrator in action for a stealth project.

Introducing Orchestrator ⚡. A single-binary workflow engine that just works™. To scale horizontally, start the binary in worker mode on as many machines as you want. Orchestrator distributes the load automatically. github.com/webstonehq/o... 👇 Orchestrator in action for a stealth project.

I needed an orchestrator, like @kestra.io or windmill.dev. 4 hours later, I have my own. Windmill benchmarks coming soon™. A single binary that ships a @svelte.dev app as a single HTML file - bundleStrategy: inline. Run the same binary in worker mode to scale workflow execution horizontally 🏎️.

Does anyone here use immich.app? I'm thinking to host it on a Hetzner machine (backup TBD)

Immich

Self-hosted photo and video management solution. Easily back up, organize, and manage your photos on your own server. Immich helps you browse, search and organize your photos and videos with ease, wit...

immich.app

So claude's new 60s "I'll do what I think costs the most tokens" timeout for questions is what I needed to finally move on. I'm fine with the Anthropic subscription, but getting tired of the CLI. Do we just use OpenCode? pi.dev? Tell me, I'll test them all 💪

That's fine. You live and learn. Nothing wrong, at any level. More AI will fix it when models get better. Too bad too many people get hurt by this nonsense.

Dare Obasanjo@carnage4life.bsky.social · last mo.

Ford has rehired hundreds of senior engineers after replacing them with AI backfired and has cost the company billions of dollars. AI adoption is blamed for Ford being the most recalled automaker in the U.S. with 51 recalls covering over 11 million vehicles in the first half of 2026 alone.

Days since I migrated a project from @cloudflare.social to @railway.com: 0 The reason is always the same: Things just work.

Apple prices increase by up to 20%. If they keep going, I'll soon be able to justify a @frame.work with similar specs. Unless of course they'll increase prices too 😂. Best thing to do is convert a rental suite into an AI data center, that'll likely yield the best profit.

Thank goodness, says the man who looked at the stock market today.

Where are the local LLMs? techcrunch.com/2026/06/22/a...

Anthropic says Claude may want to see your ID | TechCrunch

Claude's chatbot may ask to verify your age and identity "in certain circumstances," such as with a passport or driver's license, according to a privacy policy change.

techcrunch.com

It's Windows defrag all over again 😂.