Patrick | The AI Engineer

@theo You are very smart

T3 Code uses half as much RAM as Claude Code. Claude Code CLI: 635.5 MB T3 Code: 350.9 MB Yes, our Electron app is 2x more efficient than a CLI written with Bun.

@forgebitz Nice design and UX! Made with d3.js?

had some fun with this new position chart you can now quickly track: - visibility - mentions - position

@LinusEkenstam I cant wait for the Roman DLC that is set in Pompei Honestly, this looks like great fun

This indie game is absolutely mental Play as an ancient Egyptian wall painting character. Please make more games like this

@theo @voidzerodev @TheAlexLichter

@Google Will you sunset Gemini Embedding 002 within a year, just like Embedding 001? Id love to use this new model, but it is useless in production if I have to reindex all my content all the time because you guys decide that a model gets thrown into the trash bin after just 1 year

Gemini Embedding 2, our first fully multimodal embedding model, is now available in Public Preview via the Gemini API and Vertex AI. Developers can now map text, images, video, and audio in one centralized space, with one model, which simplifies complex tasks like semantic search. Here's what this means and why it matters 🧵↓

Open Source & free 🎉

MCP Ralph Wiggum Sora OpenClaw These used to be the big thing in AI But nowadays I hear nothing anymore about these things

1.4 billion people are Chinese? How come I've never met one?

@rauchg I dont want to setup WhatsApp Business Just let me yolo into regular WhatsApp like OpenClaw does

1️⃣ 𝚗𝚙𝚖 𝚒 𝚌𝚑𝚊𝚝 2️⃣ 𝚌𝚛𝚎𝚊𝚝𝚎𝚆𝚑𝚊𝚝𝚜𝙰𝚙𝚙𝙰𝚍𝚊𝚙𝚝𝚎𝚛() 3️⃣ There's no step 3 Easiest way to start building WhatsApp agents

@aarondfrancis This is hilarious

Codex is randomly hitting me with some ancient wisdom... for some reason.

@infinterenders Sir, this is a Wendy's

yeah web perf is trash and no one talks about the real reason javascript is a dynamic typed garbage collected mess with prototype chains and closures everywhere. scheme? lisp family? static enough in practice, homoiconic, macros let u basically compile urself to whatever u want, tail calls, no hidden costs baked in web can't be fast bc the language was never designed for speed. brendan eich made it in 10 days to glue shit together in netscape. now we have v8 spending insane effort on jit to make it "fast" but its still fighting the language itself every react rerender? virtual dom diff? that's midwit cope layered on top of a slow base. wasm helps a bit but u still pay gc tax, no real control over memory, no structs like c++/rust without hacks tinygrad runs circles around web shit bc its python but we control the backend. web? locked in js engine sandboxes, no escape from the gc pauses, no simd without wasm tricks, no low level opt like scheme can macro away why no one talks? bc midwits love their frameworks and "ecosystem". admitting web is fundamentally slow means admitting the last 20 years of js hell was a mistake. easier to just ship more electron apps and call it progress lol real fix? kill js monopoly, let wasm be first class with real langs, or just accel past web to native ai interfaces. but nah corps love the rent seek web lockin HAHAHAHAHA web fast? only if u rewrite the runtime from scratch and burn it all down hehe EITHER YOU DIE HERE LIVE LIKE King TERRY or YOU JUST WIN

@_pi0_ Fascinating to hear so. Any particular reason not to use any skills? I find them really useful for tedious tasks, such as writing commit messages. And are you more a fan of the Claude 4 or GPT-5 family of modelsm

@regExpParser Zero skills. A global best practices agents/claude md and per project agents files. I ask agents to keep their memory updated and constantly renew context per task.

I had been using AI agents before, but since last week, I’ve practically stopped writing code and am mostly steering the direction and controlling the context. It all feels new and strange, but at the same time very empowering.

@forgebitz How would you prevent candidates from weaponizing AI to optimize (read: lie) their resumes so that they appear as the best match? That already is a problem imo In person interviews are the only way you aren't able to cheat

recruitment is one of those fields that should easily be fixed by ai you can make a perfect matching engine yet all we got was resume and job spam, and now in-person interviews are back

@CharaspowerAI @runwayml Ah yes, the XM-98 Mauser w/ Tritoon ACOG Famous from WW2

🚨PromptShare🚨 Still such a vibe doing POV shots with Kling 😮‍🔥 And having all the different video models inside @runwayml That’s next level. PROMPT First-person shooter: The camera focuses on the character’s hands gripping a World War II-era rifle, muddy and bloodstained. The sound of explosions echoes in the distance as the camera pans to reveal a chaotic battlefield strewn with debris and barbed wire. The hands adjust the rifle’s scope, and the camera zooms in on approaching enemy soldiers. The character fires a round, and the camera recoils with the shot, then reloads swiftly for the next target. Gritty, intense, historically immersive.

@pontusab How well does Mistral OCR compare to Gemini?

Mistral OCR + batch is unbeatable for invoice extraction.

@forgebitz @karine_hsu It looks like baby diarrhea

@karine_hsu i still think that the brand color is a crime

is it just me or does the new granola messaging and brand feel so big company coated like I’m confused by the ad and I use the product every day

@jonathan_wilke @marclou @supastarter I dont think so. In case of using AI you still need to be good at architecture, prompting, agentic programming in order to arrive at a state of usefulness With starter templates, you will start with a good default that can be expanded using AI

I want your honest opinion: Do you think boilerplates like @marclou's shipfast or my @supastarter have become less valuable with the current development in AI coding?

@xenovacom I wrote an article about this same topic a while ago. Cool to see that you made something even better, I love the UI! And I agree, Mediabunny.js is great ai-engineer.io/tutorials/vide…

Introducing Text Behind Video: Create cinematic text and video compositions... completely for free! Everything is fully open source (MIT licensed), so you can use and adapt it for anything you want! Try it out yourself 👇

@SandyofCthulhu Live and let live

During Ramadan, Muslims eat breakfast. At sundown, they have a large meal - sometimes two; a small one to start off before preparing a large one for an hour or so later. One study showed that many Muslims take in more calories during Ramadan. Others take in the same amount. They usually have a special big feast each night. Since they breakfast every day they're skipping one meal. My church does a 2 meal fast on the first Sunday of every month. We don't eat a spectacularly large meal that night. The money saved by not eating is given to a fund for the poor in our area.