AI agents can code. They still ship slop UI.
So I built a directory of 100+ design skills for Cursor / Claude / Codex — taste, motion, typography, anti-slop UI.
What’s in it:
• Anti-slop / taste skills (Hallmark, Impeccable, Taste Skill…)
• Motion (GSAP, spring, scroll storytelling)
• Visual systems (type, color, shadows, layout)
Browse, pick a skill, drop it in your agent.
Didn't expect my LinkedIn content to hit ~1M impressions and reach 250k people over the past year, growing 270% YoY.
I'll be sharing even more openly this year about AI, engineering, building things.
Here is my LinkedIn: https://t.co/nzzlBURbf5
After testing many models for our coding agent on https://t.co/WVCknX6n9n, GLM 5.2 (@Zai_org) is the first one I can use daily for most tasks. OSS models are catching up fast.
https://t.co/4pBlHhfeZU
A simple note-taking app with a few AI features. No fancy UI, no complex integrations, no 100+ features you'll never use.
Everything runs right in your browser, maximizing privacy and security.
Think Apple Notes on the web — with a bit of AI.
Built on https://t.co/CAnh6LUC0k
Something that's been stuck in my head this past week: AI labs are living 3–6 months ahead of the rest of the world.
While most of us are switching between Sonnet and Opus to save on cost, the engineers inside Anthropic are probably already running Fable 5 as their default — before most people have even tried it once.
And with the best model in hand, it's not just that your execution gets faster — your ambition grows too. You start building more advanced solutions because suddenly they're within reach.
The gap compounds over time.
Refactored the https://t.co/WVCknX6n9n landing page with Claude Fable 5.
It's ridiculously good at frontend and UI/UX — basically rescued my weak design skills.
Shame we can't access Fable 5 anymore.
Fable 5 (1M) + Ultracode (Dynamic Workflows). 12 agents running in parallel in the background, consuming a total of 700k context.
I don’t know if the code works or not, but it’s beautiful.
Trying flagship model on hard tasks and burning tokens is my new hobby now.
My current personal AI development stack for building @hi_michii
- @conductor_build for running agents in parallel
- Opus 4.8 + Fast Mode (Dynamic Workflows for hard tasks) for planning
- GPT-5.5 + Fast Mode for implementation
- Opus 4.8 + Claude Code for code review
I have been reading @chipro books since I was a uni student, and finally met her in person. We had a 1-hour walk and conversation about AI and startups. Great to see Vietnamese thriving, especially in SF 🇻🇳
Building software is easier than ever.
Building a business is still really hard.
So I'm building Michii — an agent team that plans, builds, AND markets your product. Not one AI that codes. A full team: product, engineering, growth, analytics — in one place.
You stay in control. Agents propose, you approve.
Check it out: https://t.co/CAnh6LUC0k — would genuinely love feedback from anyone who's tried.
Never thought my first workday in the US would be at an accounting firm in suburban Colorado, where all you can see around you is mountains. Scenes from Breaking Bad or Better Call Saul — now I’m seeing them in real life 😶