I am thinking about https://openclaw.ai/ the open source agent that you can install locally that will proactively manage and execute tasks on your local computer. This has been the buzz for the last three or four weeks. We are at one of those turning points in our use of AI, we can move from interacting with passive, reactive tools to setting up systems that have agency and can make changes proactively. It is super exciting and frightening to think that you can install a worker that will do your work for you. My personal experience of active agents with my tests on browsers like Perplexity's cursor is that it's alarming when your instructions can be misinterpreted and extremely dangerous to do anything on your personal or work computer. Like most people, my recommendation would be to set this up on a separate, clean server or PC. I see people are buying Mac minis for this, but there's not much point in setting up an agent like this in a vacuum. That is the safest thing to do, though. I don't think we're anywhere near granting agents unsupervised access to our email inbox, files or more. The vast majority of us do not have the guardrails or monitoring, to do this safely right now. I'm really curious to see if there are any actual use cases for Open Claw out there. The only one I've seen is for tidying up your desktop. Now, Claude Cowork, on the other hand, seems to be set up with guardrails, with restrictions in place that make it safer. You can limit which directory it can work in on your local computer. Cowork seems to unlock the magic of multitasking and stacking sequential tasks, which is undoubtedly huge. See this post for a great starting point: https://lnkd.in/dnjrmHUp That being said, we are really moving into a new area, and the people who have an understanding of how to restrict and protect sensitive information and processes and keys from agents are other people who can be most adventurous.