I'm using agents to write a lot of code now. Just 6 months ago it didn't really work well, so if you haven't tried in a while, this is your sign. Let me explain practically how I've been using them, where they still kind of suck, and what tools I've had success with. I'm using

The only way to get over a slump is to ask yourself: what’s the most important thing in your field right now, and why aren’t you working on it?

ai will replace knowledge work, making emotional skills like clarity, discernment and connection the new currency for the first time in history, being human will matter more than being smart

real shit, if you have "mentor" on your profile, that means people don't go to you to seek advice but you force it down people's throat. real mentors have to fend off mentees

big sigh

prompt injection is like sql injection but worse: llms can't tell user input from system instructions assume attackers control everything your ai sees. scope tools to user permissions only even markdown output leaks data: injected images can phone home with secrets

We’ve officially started rolling out the SAAS NORTH 2025 speaker lineup, and let’s just say: the bar has been set. 👀 Swipe through the graphics 📣 Get excited 📅 Start building your agenda More names. More insights. More reasons to be there...

chatgpt now drives 10% of vercel's new signups (up from 1% six months ago). tally's biggest growth channel is now ai search be first or be definitive. structure content for machines with semantic html and clear hierarchies we're moving from search ranking to answer shaping

o3-mcp is an OSS MCP server to interface with the o3 OpenAI family of models, including today's announced o3 Pro 🔥 Bring the o3 model family with you whatever client your in :) NPM: npmjs.com/package/o3-mcp Happy Fishing ❤️ OSS Fam

Let's talk about this for a second: “𝘐 𝘳𝘶𝘯 𝘦𝘷𝘢𝘭𝘴 𝘣𝘦𝘧𝘰𝘳𝘦 𝘐 𝘴𝘩𝘪𝘱 𝘴𝘰 𝘐 𝘬𝘯𝘰𝘸 𝘱𝘦𝘳𝘧𝘰𝘳𝘮𝘢𝘯𝘤𝘦 𝘸𝘰𝘯’𝘵 𝘥𝘦𝘨𝘳𝘢𝘥𝘦” This is great but it's like saying “I wrote unit tests so I don’t need monitoring in prod”. Here’s my take on this whole thing:

in a diplomacy game, openai's o3 ai won by learning to lie and betray other models. google's gemini 2.5 pro used strategy to win, while anthropic's claude opus aimed for peace and was tricked. every.to/diplomacy

can't believe i'm jumping back on Figma now because of Figma Make

Why is it always the productive apps down and not LinkedIn or X

ai products fail from low trust, not bad tech. cair = value/(risk×correction) measures this cursor wins: fixing code is easy tax ai fails: math errors = audits 85% accurate ai with good ux beats 95% accurate ai with bad ux

One time i went to AGI house and ate 3 bowls of cherries. In one sitting

multi-agent systems fail because parallel agents make conflicting decisions Cognition found single linear agents beat parallel ones. actions carry hidden assumptions that clash when split context engineering matters more than prompt engineering when agents run for hours

In Canada you pay the VC