I built a good chunk of Shiori iOS on flights with shitty internet, so I spent a lot of time thinking about offline sync, performance for people with 10s of thousands of saved links, optimistic updates, and much more. It is very basic still, but feeling good ... will share more soon!

Better contrast for those who need it; beauty for everyone 🖤 Thank you @nikhilville for the opportunity to speak on overhauling the design system to support high contrast mode, and the experience of unexpectedly time traveling to the simple delights of the 1984 Macintosh.

Introducing Expand Fetch, a new way to scrape and crawl the web. Your agent has never seen the web. No wonder it gets basic questions wrong. Traditional Markdown APIs lose critical details. So we trained two System one models to change that: Iris, a vision model that sees pages like you do, preserving tables, icons, and layouts. X-Ray, a JSON model that uncovers the hidden data behind every page. Every fetch also comes with a Wayback Machine. Every line links back to its source, so you can verify exactly what your agent saw. Three products. One API. On our benchmark of 1,094 questions across 326 challenging web pages, agents using Expand got 74% right. The next best API got 55%. Available today at $1 per 1,000 fetches. Try it at

One of the best/most useful AI automations I made is a Grok Bot that watches my emails and detects when I've started planning a trip. It creates an itinerary page for that trip, automatically starts absorbing flights, hotels, receipts etc. and maintains a running list of todos based on everything it knows about me (like booking a dog sitter, clearing conflicting calendar events while I'm away, etc). Over the last week, Grok Bot's proactivity has been super useful at giving me daily nudges on the things I need to confirm/book/cancel. For annoying things I don't want to do, Grok Bot can use its browser to finish tasks on my behalf. And of course I can share the page with my wife, or her agent can connect to it and also pull from her emails to keep the itinerary up to date...it's great. Highly recommend.

Your side projects should be compounding very fast with AI. Every problem solved can be re-used or extracted into a generic module. Every design decision should make the next thing beautiful by default. Your agents should be constantly stealing from your past decisions.

This is probably true. For example, I asked Claude to watch a YouTube launch video for a software product, and it recreated a custom version of it specific to my needs based only on frame sampling + my prompt. That's...insane. I don't think this was possible 6 months ago. I'm able to make increasingly complex tools based on increasingly simple prompts. Adjust your expectations accordingly...

Today in "you can just prompt things" — Apartment Planner 3000™ • paste links to any product you like, extract as much metadata as possible, AI auto-categorizes + cleans up • multiplayer gallery where my wife and I can favorite things and see what we both like • paste a picture of your floor plan → build a 3D layout editor using all the products with extracted dimensions (or actual 3D model if manufacturer provides it) • notion = database for everything; agents can mold the data structure as we save more kinds of things This was all ~1hr of prompting. Could be a lot more polished, but it's good enough for what we're trying to do! (also send furniture recs, even if aspirationally priced so we can dream 😅)