Anthropic launches Claude Duet, a proactive agent that runs tasks in the background and waits for approval on sensitive actions

Cool, but my worry is these things dig through every integration you hand them and force relevance where there's nothing real. More integrations and more context isn't obviously the answer.

If your threat model is strict, the only real answer is going fully local. 24GB VRAM models are usable now, just not at hosted-product level.

What are people actually using these assistants for? I still see demos of reservations and email triage as the main use cases.

I run mine over a tailnet to handle CLI work on my machines, routine repo work on GitHub, and periodic grocery orders. Anything loopable is a waste if you're not looping it, and I just say yes each time.

These companies are shipping their org chart. Why isn't this native to Claude instead of a separate app, and why Opus instead of the latest model?

This is a direct competitor to Dot and Muse. I've been using Muse lately and it's surprising how little memory management you have to do when it's one long thread.

Instinct being worth billions with no compute, no model, and no revenue looks a lot like the Clubhouse moment to me.

I think the model is the differentiator. Muse's model felt weak to me, and if Opus does better, that's all I need. Being locked into one model family doesn't bother me much.

Dots was worse than I expected for coding, but it's been good at watching my channels. It surfaced buried emails that actually mattered.

Grok Bot caught an AWS bill on a forgotten account that was pushing 50 bucks a month, so it's already paid for itself for me.