Muhammad Ahsan Ayaz 🦋 @codewithahsan.bsky.social

I built a Pomodoro timer for streamers because I was tired of alt-tabbing during my coding streams. 60 seconds to set up. No plugins. No downloads. Here's how it works 🧵

Announcing... Waraq (ورق) - Translated as 'leaf', or 'page' ! Most of my Islamic books are scanned PDFs..every page is a photo. On a Kindle that's pinch zoom, scroll repeat. So I built waraq. It turns a scanned Urdu, Arabic or English book into a Kindle book that actually reflows, with real page numbers and linked footnotes. The part I care about most: every Quran verse gets checked against the canonical text and corrected. Across 20 volumes of tafsir that was 23,434 quotations It's made for your AI agent. Install it, run waraq setup-agent, then tell Claude Code(or Codex, Antigravity) "convert book.pdf into a Kindle book" and it takes it from there. npm i -g waraq Free to use. Built with Claude Code doing most of the typing. Looking forward for folks to try it on their own shelves :) Readme in the first comment 👇 PS: Animation and logo created by @Google Gemini 4 Argon! Inspired by @fofrAI 's experiment.

AI models are trained on the same games, making benchmarks unreliable. Testing on obscure street games like Pittu Garam reveals true reasoning vs. memorization. Build Flappy Bird? That tests memory, not intelligence.

The 'price per million tokens' metric for AI models is misleading. Focus on 'cost per completed task' for a true understanding. We're diving into issues with Kimi K3, Claude Fable 5, and Gemini 3.8 Flash.

Made 3 AIs Build a Game They've Never Seen | Kimi K3 vs Claude vs Gemini youtu.be/cQlr4Zs2qWQ

Speaking at AGNTCon + MCPCon tomorrow about how I built an orchestra of AI agents for over 5,000 developers. If you’re there and see me, come and say hi. 👋🏼

#BuildWithAI Live: How to Start AI Projects in 2026 xitter.gq/i/broadcasts/1pKdRDPgX…

To be honest. I don’t see a reason why this is bad… especially given the acquisition. And I’m pretty sure the discussions behind this would’ve been crazy

Was trying to go through some features of a backend hosting provider. Looks like I need to consult a doctor. Thanks, Sonnet 5.

Last night I built a tool that strips burned-in captions out of a video. Runs Llama locally, finds the caption regions, re-renders without them. About 90-95% accurate...it still leaves a stray box here and there. It took me one evening. And that's the part that bothered me. Starting a project used to cost something. You'd think about it for a week because scaffolding it was a day of work you'd rather not waste. AI removed that cost entirely. What it did not remove is the cost of finishing. So now I have tools. Lots of them. Most of them at 80%, which is the exact percentage where a thing is useless to anyone but you. It's impulsive shopping. Same mechanism...the friction that used to make you pause is gone, so you buy, and the thing sits in a cupboard. The fix I landed on is putting the friction back on purpose. A git hook that stops me creating a new repository and asks me a few questions first. If the idea is genuinely fresh and I can't answer them, I don't get the repo. One rule that matters: if I hit a guard I can't get past in an emergency, that guard gets uninstalled the next day. A guard you route around silently is worse than no guard, because now you're lying to yourself about having one. This is the same thread as my TEDx talk at KTH...we keep optimising for how fast we can start. Curious whether anyone else has built something like this for themselves.

Let’s go 🚀 this is a golden opportunity