Sorry for being an idiot, but what is an agentic OS?

Agentic OS is a buzzword that's meant to imply that the OS is (or has) an AI agent doing useful things for you in the background without you explicitly asking it to do those things (ie an agent working for you). For an agent to be useful, (they say) it has to know and learn everything it can about you, your life, your friends, activities, contacts, work, and so on.

The tradeoff is pretty extreme though. Everything you do on the PC is watched, analyzed, catalogued, and retained by MIcrosoft (and possibly whoever they choose to share the info with, which is likely every government that asks). The features that do this are generically called client-side scanning and Microsoft has a few specific variants you can read about called Copilot Recall, and Copilot Vision.

Windows 10 had a better kernel than 7. Unfortunately, that kernel was packaged with the rest of windows 10.

"Agentic" is the buzzword to distinguish "LLM will tell you how to do it" versus "LLM will just execute the commands it thinks are right".

Particularly if a process is GUI driven, Agentic is seen as a more theoretically useful approach since a LLM 'how-to' would still be tedious to walk through yourself.

Given how LLM usually mis-predicts and doesn't do what I want, I'm no where near the point where I'd trust "Agentic" approaches. Hypothetically if it could be constrained to a domain where it can't do anything that can't trivially be undone, maybe, but given for example a recent VS Code issue where it turned out the "jail" placed around Agentic operations turned out to be ineffective, I'm not thinking too much of such claimed mitigations.

Wild guess:

Log everything the user is doing. Have clippy interface prompt the user to take some work off their hands. Do some web searches, start storing a dossier about the 'project'. Give the user a rough outline to complete their project based on a trained llm and some web searches. Ask the user if the outline looks good. Ask the user if they'd like some help completing some of the steps. Burning through tokens the whole time, storing telemetry with 100% knowledge of what they user does/wants to do. Selling that exact data to project management software companies and companies that write middleware to do this work. Bind everything together into a virtual notebook where users can return to any content at any step.

It's some AI crap they try to push on us.

This is really just a guess but... I think "agent" in this context means a personalised AI.

Training gen AI models requires huge amounts of resources. Its not practical to train an AI for your personal use.

Creating an agent is something like, taking an existing model, asking it to keep your entire browser history in mind while you ask it to do your homework.

IMO its actually one of the big limitations of gen AI, but somehow the word is supposed to mean the opposite. As in, the current approach has reached a dead end requiring exponentially more resources for less and less improvement. So because we can't make a model that just knows or learns everything, we have to make agents that know lots about specific things.

I mean I would love a linux distro that had a local agent that could answer queries. I view it as sorta like graphics compared to command line but if it needs networked resources outside of it doing its own searches. No thank you.

Which is why Windows 7 is the best operating system microsoft has, and seemingly will, ever produce.

hear hear!

Win 7 sucked too.

Win XP was the beginning of enshitification.

You just have nostalgia.

Agent work can be pretty magical. I've been using cursor recently and the fact that it can just execute commands on your PC means you can just tell it to do something and it does. Troubleshooting as it goes