Every new AI feature, turned into something you can build.
I am Omer Bin Saleem. Claude ships a skill, OpenAI ships an agent, a tool gets a new trick. I try it the same week and show you the steps, the prompt, and what it is good for.
Every week, a hundred new AI tools ship.
I try them the same week.
And keep only what works.
Most AI content shows you a demo. I show you the steps, the prompt to copy, and what it is actually good for, because I use every one of them in my own work first.
Hi, I’m Omer
I build with AI every day, and show you how.
Frontend lead by day. I run a small team of AI agents for my own work, try every new AI feature the week it ships, and turn the useful ones into one short hack a day: the steps, the prompt, and what it is good for.
Omer Bin Saleem on InstagramOmer Bin Saleem on LinkedInGitHubYouTubeWhat I buildMy Claude Code setupMore about me
Inside the hacks
Slides from every build.
Today’s hack
A 94 second launch film from one HTML file
No editor was opened. One HTML file of 436 lines, 14 script lines read by a local voice model, every cut snapped to a 112 BPM beat grid, 11 phone recordings inside a CSS phone, 2,838 frames rendered in 62 seconds.
3 min read →Latest
All hacks →3 print CSS traps that broke my PDFs
Our PDF preview took 11 seconds and still did not match the screen. Printing the same HTML the editor renders, with headless Chromium, took it to 2.5 s cold and a 30 ms render. Three print-CSS traps cost a round each.
3 min read →The feature walkthrough that floated into empty space
An in-app guide fired on mount, tried to anchor to an element inside a conditional section, found nothing, and drifted. The fix is one check per step: is the target in the DOM right now? If not, skip it.
2 min read →Put a 0.25 second judge in front of your AI agents
Big models do the work. Around that work sit hundreds of small yes-or-no questions a day. A small judge model answered 116,097 of them in 20 days, in a median of 245 ms, and only acts alone above a confidence gate.
4 min read →My AI tester took 9 minutes and proved nothing. A week later: 1 minute, one real bug.
When an agent disappoints, most people swap the model. The fix was what it knows before it starts: a steps file instead of a script, a playbook, and a lessons file it writes itself.
3 min read →Lead a team of AI agents, not one chatbot
One lead you talk to, narrow agents that each do one job, and a gate in front of anything you cannot undo. This week the team got its own lanes, a QA runner, a rule book and a voice.
5 min read →Your team’s questions already have answers. Give them a command.
Every week the same four questions hit the team channel. The answers were already in Jenkins, GitLab, the gateway and Jira. Four read-only commands now answer them in seconds.
3 min read →This week in AI
All trends →- Loading this week’s AI news…
Hacker News, OpenAI, Google DeepMind, Hugging Face, GitHub and Simon Willison, refreshed every six hours.
Free guides
All guides →Verify the call
The one-page checklist against deepfake CEO fraud: the code phrase, the saved-number rule, the red flags on the call, and what to do in the first hour if money moved.
How to get it →Patch php2shell with one prompt
The fastest safe path from "am I exposed?" to "patched and restarted": the fixed PHP versions, a patch-only prompt for your coding agent, the restart step everyone forgets, and the stopgaps that buy hours, not days.
How to get it →Claude Motions: a launch film from one HTML file
The prompt guide behind a 94 second product launch film that no editor touched: the brief prompt, the script rules, Kokoro voice, a librosa beat grid, footage in a CSS phone, and the render that took 62 seconds.
How to get it →Lead a team of AI agents
One lead, narrow agents with one job each, gates in front of anything you cannot undo, and rules that keep themselves healthy. With prompts to copy.
How to get it →Put a small judge in front of your agents
Where a 0.25 second judge model fits next to big models, the five rules that keep it safe, my numbers from 20 days, and three prompts to add one to your codebase.
How to get it →Researched
All research →4 in 10 GitHub pull requests now start with an AI agent
More than 40% of pull requests on GitHub in September started with an AI agent’s first draft.
1 source →An AI lab’s $104B loss that is really $1.3B
Anthropic’s updated IPO filing shows a $104.0B net loss for the nine months to September.
2 sources →Pakistani freelancers earned a record $1.76B. That’s about $53 a month each.
Pakistani freelancers earned a record $1.76 billion in FY2025-26, up 78% in a year.
5 sources →