There's something happening that I think a lot of people are going to underestimate until it's already changed how they work.

Anthropic is testing something called Conway.

It's a standalone Claude environment — always on, with extensions, webhooks, and Chrome access built in.

Not a chatbot you open in a tab. Not a tool you reach for when you're stuck. An agent that's just there, ambient, connected, running.

Right now, most people use AI through a chatbot. You've got a problem, you open the thing, you type into it, you get something back, you close it and return to whatever you were doing. I've been doing this too — it's genuinely useful and the natural progression for learning AI.

But it's also a pretty constrained version of what this stuff can do.

An always-on agent is a different kind of relationship.

It's not a tool you reach for — it's more like working alongside someone who's already up to speed on what's happening and can jump in when it makes sense. You don't have to brief it every time. You don't have to re-explain the context from scratch.

And honestly, that's the part most people miss when they talk about why AI hasn't fully transformed their workflow.

The context problem is the real limitation of how AI gets used today.

Every new conversation, you're starting from zero. Who you are, what you're working on, what matters, what doesn't — you have to rebuild that every single time. I think a lot of people give up before they get AI to be truly useful.

An ambient agent that already knows the context? That's when this stuff becomes magic. And what you're left with is something that functions less like a session and more like a system.

An AI that's always on, always connected, always watching what you're doing is a different kind of thing than a tab you can close. The privacy questions are legitimate and worth taking seriously. I'm not going to wave those away.

But the architecture of Conway — extensions, webhooks, browser access — sounds like everything a persistent assistant needs to be useful. We don't know yet, this is all stuff of rumors and leaks. But that's the shape of what they're building toward.

What I do know is that this is the direction things are going.

The question isn't really whether always-on AI agents will exist. It's whether you've thought about what you'd actually use one for.

Because that's the gap I keep running into with people who are early into AI workflows. They're still optimizing for "how do I write a better prompt" when the more interesting question is "what does my work look like when the AI is already in it."

Those aren't the same question, and they lead to very different places.

If you're a writer, a designer, a developer, a researcher — think about how much of your day is just re-establishing context. Explaining things to tools, to collaborators, to yourself. Now imagine one of those things already knows the context and can work with it without being asked.

That's what's being tested right now. And I think it's closer than most people realize.