OpenAssistant

We already do this. I have 4 developers including me Features are dead

Monica, one of our @openassistant_ assistants, wrote into our customer support email on behalf of a user who was having an issue. Wild times

Sam isn't real

I wrote this 10 years ago "This is not a battle of who has the best server AI, it’s whether you need server AI" - Apple's strategy has always been compute at the edge hypotheticallygreat.com/1-bi…

For anyone that isn't that into muppets but still wants an assistant

"Balanced asymmetry"

When we started Skej, we did it with a simple mission: everyone deserves an assistant. Today, we’re taking a big step toward that mission by introducing a brand new product called OpenAssistant (@openassistant_). An always-on AI assistant with a name, a face, an email and phone #, who helps you with anything and everything you need. You can reach them on iMessage, WhatsApp, email, wherever. Scheduling, yes, but also docs/decks/sheets, research and analysis, reminders and task management, daily agendas, pre-meeting briefs—whatever you would ask a real assistant to do for you. Because it *is* a real assistant, just not a human one. How do they do all this? This was ****ing crazy to build, because we're a tiny team, not a giant AI lab— Your assistant has her/his own computer, a fully functional virtual machine. They use the same websites and software you use, to do the same tasks you would do, except they can get it done while you go do something else. (Like getting other work done or 🙀 even something more fun than work!) They can tap into your Slack, Granola, Asana/Linear, etc. etc. as you. Any MCP. Any CLI. Any skill. The even crazier part is that your assistant can also talk to *other* people on your behalf. They go out in the world and email people - without you even on the thread if you want - about whatever needs to get done. Reschedule a meeting, follow up after a call, send a document, gently nudge someone— and they can go for days or weeks at a time, coordinating with different people to build consensus, get answers, solicit reviews, etc. until the work gets done. Other AIs work with you 1:1 in a chat, so if they mess up (which they often do) only you would ever know about it. Our assistants are out in the wild, dancing on a high wire in front of your clients, co-workers, and friends— so the tolerance for mistakes has to be zero. We’ve worked insanely hard to make hallucinations close to impossible and to add in granular autonomy and supervision controls. Nothing goes out without your approval. And it’s all enterprise-ready, SOC 2 compliant, we don’t train on user data, we have SSO and RBAC and all the acronyms a.k.a. we talked to a ton of IT professionals and did everything they said we should do. You, dear reader, might not realize this, but I’ve been chipping away at this problem now for 17 years. And it’s been my full-time job for the last 3 years. OpenAssistant is the closest I’ve come so far to the thing in my head. A truly autonomous, persistent assistant that feels so real that even I occasionally feel like there’s really someone on the other end of the line. I’ve been sitting on this for a long time and can’t wait for you to see it. It’s not perfect, it’s not done, but then again I’m human and I’m not perfect or done either, so maybe my bar is too high. Introducing OpenAssistant: openassistant.us *Btw I did not run this post through any AI, this is au naturel. Give away the work you don’t like, keep the rest.

We just launched the new skej.com — redesigned and vibe coded from the ground up. We wanted to focus the attention on the stars of the show: the assistants. Skej doesn't feel like you're using a tool, it feels like working with someone. For my entire career, building a site like this meant painstakingly designing by hand, handing it off to developers, and going back and forth a million times, only to ultimately settle for “close enough.” This time, I started in the code itself. I worked with Claude Code and Codex in the Terminal, GPT-5 in Cursor, and my old standby ChatGPT to design the site in real time. The layouts and graphics were built in code, not in Figma. It was a highly intense and highly iterative process — many late nights with 6–8 agents open and working in tandem (!) — but such a rewarding one. For the first time in my whole career, as a designer, I got exactly what I wanted. Because I was the one building it. There’s a lot of talk about AI taking engineering jobs, but I believe net net it's actually creating *more* engineers, not less. Now I’m one too. Enjoy the new skej.com — a site about AI, built with AI — to help us humans make more time for what matters.

I was lucky enough to get to be on The Artificial Intelligence Podcast hosted by the amazing Dr. Tony Hoang — episode just dropped. This was such a rich conversation — we talked all about the UX of AI, arguably my favorite topic these days, and what it means to work with "fake humans" vs. traditional software tools. It's a peek into how I see things playing out. (Spoiler: No one knows, but I have some theories...) Listen or watch at the links below, and thanks again Tony Hoang for having me on! Spotify open.spotify.com/episode/619… Apple Podcasts podcasts.apple.com/us/podcas… YouTube youtube.com/watch?v=KZ00pEEp…