AI-built sites all look the same. I built an open-source workflow that fixes that. loom gives your coding agent distinct design directions, honest screenshots, and a reviewer that sends the work back until it passes. Works in Claude Code, Cursor, Codex and Antigravity 🧵

Applying for this one, and their application asked for a quick video on something I learned recently. Made this one about noticing Claude Code @claudeai uses words like "cogitated" and "discombobulated" in its reasoning, turns out that's the model describing its own confusion-to-understanding process as it works through code. Enjoyed making it enough that I want to turn "something I learned recently" into an actual series. Feels like good practice for growth work too, taking something dense and making it click for someone without the context. Video below.

Most subscription apps are built to help you cancel things. My problem is the opposite. I want the charge to go through. I just forget to fund the card before it does, and then Claude declines mid-task or Netflix dies on a Friday night. So I built Rekona. It tells me 7, 5, 3, 2 and 1 day before every renewal, and on the day itself. Enough warning to actually move money. Not on the App Store yet. If this is your problem too, reply or DM me and I’ll add you to the first group testing it.

How much of your code was written by AI lately? Mine too. Here's the problem nobody talks about: bugs that look fine, pass review, and ship anyway. So I built Syntinel. An AI that reads your codebase like an attacker would, SQL injection, hardcoded keys, auth bugs, prompt injection, then writes the actual exploit and the fix. One command. Real security audit. Not public yet, but here's a first look 👇 Comment "SYNTINEL" if you want in when this goes live.

Teaching a model to shop by sight 👀 Building visual product search for Zetix. Snap a photo, find the exact item. Off-the-shelf CLIP only hit 55.5% top-3 accuracy, so we're fine-tuning our own on real product images. Watching the loss curve fall live... 3.25 → 1.16 and dropping. All training locally on a MacBook

Interesting side effect of fable 5 being this good on claude code: opus 4.8 now feels like a downgrade. hit 95% of my session after hours of work, swapped models to manage usage, and the quality dip was immediate and obvious. didn't expect the gap to be that visible in real work.

Proud of the team on this. The tedxakure build was one of the cleanest projects we've shipped. excited to see this year scale even bigger.

Opus 4.8 first impressions, chat only: responses sound more technical than 4.7, and the structure feels disconnected from what i actually asked. switched back to 4.7. Code is where the benchmark jump is (+5pts on swe-bench pro), so the real test for me is there. running it tomorrow. no verdict yet. one hour of chat isn't enough.