Josh Merrill

The blank prompt box is the wrong starting point for personal AI. 🧠 It assumes the human knows what to ask, what context to include, what history matters, and what the AI is supposed to remember. That is a lot of unpaid labor before anything useful happens. 🧾 Most people do not want to become prompt engineers for their own life. They want something that already understands the shape of their week, the people they care about, the loose ends they keep forgetting, and the promises they made in passing. 🔗 The interface should not begin with “what do you want me to know?” It should begin with “I was paying attention.” ✨

Merrill’s Law: The closer someone’s point of departure is to their destination, the more likely they are to be late. 45 minutes away? You leave at a sensible time. 4 minutes away? You start looking for your keys two minutes after you were supposed to arrive. 🫠

There’s a kind of promise we make all the time that almost never becomes a task. 🤝 “I’ll send you that article.” “Let’s grab coffee next week.” “I’ll introduce you two.” “Remind me to look at this later.” 📝 None of these feel big enough to stop the conversation and create a formal to-do. So they stay where they were born: buried in a text thread, a Slack DM, an email reply, a half-finished plan. 🕳️ And then life keeps moving. This is one of the places where I think AI could feel genuinely helpful instead of impressive-in-a-demo. 🌊 Not by asking us to manage more systems. Not by turning every conversation into a productivity dashboard. But by quietly noticing the commitments we already made and helping us keep them. 🧭 The more I work on Voyance, the more I think the real magic is not “AI that can do anything.” It’s AI that understands the small things that make us better humans. ✨

I keep hearing people say AI needs more context. I agree. But context is not one clean file you attach to a prompt. 🧩 It is scattered everywhere. A half-finished plan in email. A decision made in a text thread. A calendar invite with no explanation. A Slack message from three weeks ago. A note you wrote to yourself and never opened again. 📬 And the hard part is not just collecting all of that. It is understanding what it means. When someone texts “we should do that next week,” who is “we”? 💬 You and your partner — because that’s who “we” usually means in that thread? You and the person texting? The whole team? The group from dinner last night? Humans resolve this kind of thing constantly without thinking about it. We know the relationships, the history, the tone, the plan behind the plan. 🧠 That is the context AI usually does not have. So when I think about personal AI, I do not think the answer is simply “connect more apps.” Email matters. Messages matter. Calendar matters. Notes matter. 🔗 But the real work is stitching those fragments into a living understanding of your world. Who people are. What they mean to you. What has been promised. What changed. What still matters. That is the layer I think is missing. ✨ Not access to more data. Context.

Most software is built for everyone, which means it is built for no one in particular. Calendars are a good example. Every calendar app I have used shows the week as a grid of colored rectangles with text in them. It works. It is also the same week you would see if you were someone else entirely. AI changes all that. 🤖 When Brian and I started building Voyance, one of the first things we tried was giving every event its own piece of artwork. Not stock photography. Not a category icon. An illustration generated for that specific event, with the people who are going to be there in it. ✨ The screenshot below is a weekend I have coming up with my dad, my brother, and Brian. Those are us in the paintings. It is a small thing. But replacing a grid of text with a series of images changes how the week reads. Plans feel more real. You anticipate them differently. You remember them differently. I don't know why more calendars don't do this already. But now, one does. 😉 #AI #AIart️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️

Neither half works great alone. Together it feels like this is how it should have worked all along.

Most AI apps make you sit there and wait. You asked, now watch the spinner. 🌀 That's a chat app habit, and it doesn't belong everywhere. When the app already knows what you're pointing at, you should be able to fire off a question and move on. The answer finds you.

Every AI product today is either a chatbox or a form. Each one is bad. 🤔

On the way to D.C. for Brian’s 40th. He built the first version of Voyance because we needed it ourselves. Dinner turns into architecture. Laptops come to bed. No investors, no timeline but ours. Happy 40th, Brian.

A calendar event is just a date and a title. "Brian's 40th birthday trip (D.C.)" tells you nothing. This is what it looks like in Voyance. 401 nodes. One trip at the center. A flat calendar stores when. We're building the thing that remembers what it meant.

@socialwithaayan Moving from search to the OS level is the only way an assistant becomes actually useful. The hard part isn't the connection, it's building a memory of the people and commitments buried in those iMessage threads.

Perplexity just made the biggest jump in the AI assistant space. from search engine to Personal Computer. > connects to your local files securely > works across iMessage, Apple Mail, Calendar, native Mac apps > reads, writes, and searches your files locally > set it up on a Mac mini and it runs 24/7 in the background > start a task from your iPhone, it operates your desktop this isn't a chatbot anymore. this is an AI that lives on your machine and actually does things.

@alvinfoo Controlling iMessage from the outside is one thing, but the real unlock is an AI that actually understands the context within those threads. Most assistants still treat every conversation like a blank slate.

Perplexity just launched “Personal Computer,” an AI assistant that can run your Mac for you. It integrates with macOS to manage local files, control native apps like iMessage and Mail, and browse the web continuously in the background.

@JoshKale Integrating with iMessage is the right move, but the real challenge is making sense of the mess. Most tools just search the text; the hard part is actually mapping the relationships and commitments buried in those threads.

Today Perplexity shipped everything Siri was supposed to be 💻 Personal computer now has access to: → iMessage → Every folder on your Mac → 400+ connected apps → Apple Mail, Calendars, Browsers etc... Underneath, Claude Opus 4.7 is the brain. It breaks your goal into subtasks and routes each one to whichever of 20 models wins at it. GPT for long context. Gemini for deep research. Grok for speed. Nano Banana for images. Veo for video. Codex for code. It runs 24/7. You can trigger it from your phone. Pretty sweet design too

@SandroChess @lpolovets And it’s amazing 🤩

@lpolovets @josh_io is actually building this 🤠

If someone trained an LLM on all of my text messages, its response for every prompt would just be "sorry, I'm running 3 minutes late."

@mshuffett Super cool!!

🤯 First look at Everything AI -- think Linear but for managing 100s of coding agents / instances of Claude Code. Early demo below 📷 (filmed 2 months ago - if this blows up, I'll drop the updated version in a week)