Excited about my latest research project: AXO ("Agent Accessibility Optimization") Agents are quickly becoming first-class citizens of our everyday interfaces. And recently, major players are integrating agents directly into their tech: → Cloudflare introducing Markdown for

It BAFFLES ME how deep we are in the AI bubble OpenClaw ignores the 2 most foundational considerations of software: cost and data privacy No one in the world will pay hundreds of dollars a day to use this. And there are no security restrictions on it Yet no one cares!

If you are against adding voice chat to team video games - when there is an option to simply not join chat or mute - you are a genuine idiot and should never be allowed to access a computer again

I run this prompt every time I make a big change. And it's very simple... """ Please do a deep analysis on this PR: [Link to PR] It introduces a few key things: 1. (list here) 2. (list here) ... Can you please do the following: 1. Do the changes possibly break anything? 2.

VIBECODING TIP The most important thing to do after making a big change is to clean up after yourself. If you don't do this, you will slowly lose control of your project while burning WAY too many tokens. Simple prompt below ↓

Incredible! I joined the company almost 2.5 years ago and it's surreal how much has happened - and changed - in that time. In Crypto, and now AI, expectations are higher and more urgent than any other industry. Yet we've achieved so much in such short time. More to come 🍻

We need to bring video generation to vibecoding It's really annoying that Claude Code and Codex can't tap into Nano Banana for video generation There are so many times where I want to put a cute animation in my app and can't Agent <-> agent licensing just doesn't exist yet

We're so deep in the vibecoding shithole that people think product decisions at billion-user scale are limited solely by whether engineers can change pixels.

@Cloudflare

Over a year ago, I predicted that agents would become primary users of modern UIs We've seen this rapidly come to life in recent months with various "computer-use" tools that browse by themselves Today, @Cloudflare launched support for AI agents to request markdown versions of

AEO completely changes how you should market Answer engines don't rank content the way search does They synthesize from information they can interpret reliably. The more ambiguous your messaging, the harder it becomes for AI systems to surface you or include you at all

All the new venture-backed AEO startups are missing a very key point of this imagined future: Agents will eventually traverse the web, not humans. Simply optimizing brand search results through AI is not enough. The real moat is figuring out how agents *traverse* your product

If you own a company/business, optimizing AEO (SEO for AI) is the most important thing you can do right now What's especially cool is that unlike SEO, it doesn't take months for AI to bubble you to the top. You can take over in just a few days

Anyone else feel like models get worse over time? Maybe my expectations are just unintentionally growing but I swear they produce more error-prone code over time

For the longest time, I’ve had existential dread about working in the AI space I hate contributing to something that is bad for humanity. The singularity, destroying academia, critical thought, attention spans, etc But I’ve recently realized that AI Agents have the opportunity

Powered by @hyperbrowser, an incredibly easy infra for modern AI tooling. Was so surprised I didn't know about it already! https://t.co/PZUinfIVMP

I built an Agent race 🏁 Agents are quickly becoming the new power user of digital interfaces I wanted to see which models are completing tasks the fastest and get a peek into their brain Experimenting with a quick "computer use" race for Claude, Gemini, OpenAI, and HyperAgent

I used Claude Code for so long because it was so easy to set up in the terminal. They marketed it so well. OpenAI never really pushed the "easy" experience of Codex so I never tried it. Never even knew how to set it up. Until this app. Now it's so accessible

Using Codex after months of using Claude. I'm finding it gets things right in fewer attempts. Basically 1-shotting fixes. But Claude keeps you much more in the loop. Tells you what it's thinking along the way and allows you to step in when you need. Wish I could have both!

While today's agents may seem different, they are all "ChatGPT wrappers" under the hood with different system prompts In the near future, agents will be architected differently and harness unique capabilities to achieve specific goals That's when agentic e-commerce spawns