Anthropic just released Claude Conway as a research preview. I'm gonna talk about what this is and how it could dramatically change how knowledge work is done.
So we already knew Claude was building an always-on agent similar to opencl, partly because of the recent leak, and partly because that's just what all of the model makers are doing right now. Opencl has just radically changed the industry's view of what an AI agent is capable of doing in work. And it seems that that's what Conway is.
Anthropic's answer to Opencl. Like I said, this is still in the research preview. I haven't had a chance to play around with it myself yet, but here's what we know so far.
Conway is an always-on agent. You open an instance of Conway from the Claude app, and then each instance of Conway gets its own window. Conway can connect directly to your files, offline or online.
It can use a browser, it has persistent memory, and it can be triggered by web hooks. So you don't need to use Conway directly. It can just be triggered by any other app you're using.
And maybe most importantly, it's extensible. Anyone can build and share extensions that expand the functionality of Conway.
So why does this matter for the average knowledge worker? I own a digital marketing agency, and I think the moment that companies like ours have been waiting for, the moment where everything really changes, is when we Can fully integrate AI agents just like we would a normal team member. For this to work, the AI agent needs to have their own Gmail account, their own Slack account, their own account.
In our task management system, we use notion. They need to have access to everything our team members would have access to. And we need to be able to communicate with them the same way we'd communicate with our team members.
Now, this has all been possible since the release of opencl. Problem is that opencl has major security issues and a steep learning curve. So it's too early to say for sure.
But Conway could be that convenient, secure version of opencl we've been waiting for.