I opened đť•Ź today and saw that xAI is working on a computer use tool, similar to the ChatGPT work feature OpenAI has been talking about. That didn't surprise me much. AI is moving toward doing things for you instead of only answering questions, and I think we've been heading that way since OpenClaw came out earlier this year.

What I don't think anyone has gotten right yet is an AI that does something before you ask. Right now you have to decide you need help, open something, and tell it what to do. I want one that can notice things I haven't thought to ask about and take the next reasonable step, with limits on what it's allowed to do.

Why decisions are the interesting part

I’ve never been a big believer in standardized tests or IQ tests, there are just so many types of intelligence that categorizing them in order of priority and then assigning someone a number seems…wrong to me. What I do agree with however, is that some people are smarter than other people, that some decisions just generally make more sense than what others would choose to do.

We will most likely at some point in the not so distant future reach a point where the decision of AI is superior to the decision of a human given the exact same information and context. Some would argue we’re already there for a good amount of the population.

When it’s indisputable that we’ve arrived at a place where your AI can make your decisions about your life better than you, what becomes most important?

Trust, control, context, and data about you.

I think context and personal data are the parts I can actually start building around. That's what I've been working on with Orb - Proactive AI. I want it to know enough about what I'm doing to be useful without me having to prompt it every time. Trust and control are a lot harder, and getting those right matters just as much.

Where Orb is right now

About 2 weeks ago, the iOS mobile app for Orb went live, which is very exciting. I had been working on the app/backend for about 3 months prior to the app launch and had been using it daily for most of that time.

The app can connect to your PC and then control it with Claude Code, Codex, or Grok Build. It scans your inputs and folders every 15 seconds without an LLM call, data is logged internally and then an LLM call is made every 12 hours by default, with Orb being allowed to schedule up to four wake ups per day. Scheduling of notifications sent directly to you, such as schedule conflicts, unattended items, new news you’d need to know, works on a trigger system that Orb controls itself.

Based on things like if you open its notifications, if when you got back on your PC you actually listened, etc., Orb will learn what’s valuable versus what’s pointless to tell you, then it will proactively use whatever scope you allow it.

What I want to improve

I’ve been building largely by myself, it’s got some users but is still definitely a developing niche tool at the time being. I’m wanting to expand and make Orb more accessible to the general public, make it simpler to set up, run, and understand.

If any of the ideas you’ve heard today interest you, I encourage you to tinker around with the backend, it’s open-source, the app is free, and you can make any changes you’d like to the backend and connect to your phone using tailscale.

If you try it, I'd really like to hear what works and what doesn't. I'll be reading the comments.

Links