Turned Off By Most Always-On Agents

Ready for a singalong? Try Claude Duet. What's Claude Duet? Well, it's complicated. And a little off-key.

In late 2022, we got our first taste of generative AI. ChatGPT.

For about three years, that was enough. It was incredible, and in hindsight, incredibly limited. Claude Code was released in early 2025. Powerful tool, but made for developers. So most of us didn't get the Claude bug until January 2026.

At the same time something called "OpenClaw" was gaining popularity. This was the first iteration of what's being called an "Always-On Agent". It was a chatbot interface, but with tools like heartbeat and scheduled task that allowed it to wake itself up, take action, and be more independent than our friends ChatGPT and Claude.

The issue was. It was really hard to use.

I tried it around the same time I was experimenting with Claude. For as green as I was in agentic coding, the learning curve with OpenClaw was too steep.

But the idea stuck with us. What if AI could anticipate my needs? What if it could actually be my assistant? Those of us who played games like Halo growing up, fondly remember the character Cortana, Master Chief's AI assistant and battle buddy. (This post was not sponsored by Microsoft, unfortunately).

That's what we want. But how do we get there? We wouldn't have to wait long. The AI Hot Summer (and early Fall) was here. Featuring breakout hits like:

• Grok Bot (xAI): August 11, 2026 (early beta) - I got in 8.29.26
• Muse (Meta): September 8, 2026 - I got in 9.14.26
• Dots (OpenAI): September 29, 2026 - skipped it. Set it up 10.27.26 for this post.
• Claude Duet (Anthropic): October 21, 2026 - I got off the waitlist 10.26.26 (Research Preview)

Anyone else have agent fatigue? Just me? Fine.

For those of us not obsessing over new agents and models like they're a trendy sneaker drop, here's my unadulterated opinion about how to approach Always-On Assistants.

There are two kinds of A-OAs: Personal Assistants & AI Employees.

Both types drive conversations to you, proactively. So what's the difference?

Okay. So the market is mostly made up of Personal Assistants right now.

1. They're perfect for beginners new to agentic tools.
2. They all have tons of plugins, their own computer, and feature a primary session that goes on forever.
3. But most importantly, they focus on a single agent that you correspond with.

Not super complicated. But we're just getting started.

Muse is the most consumer friendly. Beautiful interface. Easy to approach navigation. It has a dedicated mobile app, and has seamless integration across Meta's ecosystem. I named my Muse assistant "Clio", the greek muse of art, literature, and music. Muse has a desktop app that you can give access to manage your files, just like ChatGPT and Claude.

Muse's most unique feature is the dedicated areas where you can curate a dedicated feed, add ideas that you want to brainstorm, and have Muse track progress toward your goals.

I shut mine down this week. You could say it was the day the music died.

Dots is noteworthy because it's OpenAI's answer to the Personal Assistant. I can't say I'm an expert in this one. It's ChatGPT but just more annoying. Haven't seen anything too special.

Claude Duet is brand new. It brings all of its super Claud-i-ness to a mobile interface. What's been neat about Claude Duet is its ability to build out really thoughtful artifacts using the same bones of Claude Docs and other tools in the Make tab. In fact, all of your artifacts and memory from Claude sync with Claude Duet, which is neat if you like that kind of thing (more on that in a second).

Claude also gives you the ability to manage action items and tasks in a dedicated tab.

Both Claude Duet and Muse give you the ability to have side conversations. Muse calls them Side Chats. Claude Duet has channels, which makes it feel like Slack (in a good way).

The thing I HATE about Claude Duet is that it does not connect to your computer's local files in any way.

I haven't shown off my local architecture and tech stack yet, but if you know me, you know that I'm obsessive about running my agents through my local machine.

It's not a problem, it's context engineering. All the cool kids are doing it.

I was so angry last night that Claude Duet couldn't access my computer that I setup a dedicated handoff environment for it to talk to my Grok Bots in a Notion Doc. It now can ask Grok questions, to a dedicated "Claude Duet Ambassador" (very Diplomat season 4 inspired), and Grok goes on my computer and gets the answer and brings it back for Claude Duet to use.

It's a terrible system.

Everyone I work with knows I'm a Claude fanboy. I typically am running non-stop sessions in Claude Cowork and Claude Code.

But so far, I hate Claude Duet, because it stripped me of all the benefits of what I loved about Claude: a high-powered agent that can manage files on your computer.

So what are AI Employees? It's the Shangri-La of A-OAs.

Grok Bots are different because instead of a single personal assistant, you can have as many as you want.

It's beautiful.

Think of it this way. What's better? A single personal assistant or an entire org chart of agents focused on tasks, proactively managing any designated workflow you give them.

It's like running a company.

Personally, I'm running about 30 agents across different tasks running from research, financial analysis, house management, planning, and communication management.

Grok has full access to your files, like Claude Code, and I've been really impressed with its ability to manage tasks reliably. On top of that, Grok Bots can spin up Cursor agents for code review (which I find myself doing more and more of daily) or other tasks whenever you want.

One of the main benefits from having a setup like this, is that agents can work together to accomplish tasks while managing specific context and specialization. This gives you more surface area to screen for task weaknesses. It's the same reason why we have teams in real life. The more the merrier, as long as you have clear expectations for roles and behaviors.

Now AI Employees are not for everyone. It's not a leap to say it's much harder to manage 30 people than it is to manage 1.

Agents are no different. So which do you pick?

For those of you at either end of the bell curve:

• If you're doing serious agentic work 24/7. I'd go with Grok Bots.
• If you're super new to AI Assistants and don't use them professionally, go with Muse.

For those of you that are working professionals, and have some experience with AI, your choices are:

• Do you like working on files on your computer? OpenAI Dots.
• Do you like working on files in the cloud? Claude Duet (when you can get access).

At this point, all Always-On Agents are pretty fantastic. The only people that will tell you otherwise live in San Francisco. You know what I mean.

You cannot make a bad decision beyond these preferences.

But as always, prove me wrong and I'll recant. I would be more than happy to chat about your experience with A-OAs and other AI tools. As much as I live in this world 24/7, it's moving too fast to ever claim that one has a perfect grasp of what these agents can do.