Personal AI agents compared: Instinct, Muse, Grok Bot, Claude Duet and ChatGPT dots

What is a personal AI agent? The short answer

A personal AI agent is an assistant that keeps working when you're not talking to it: it watches the apps you connect, does tasks on its own cloud computer, and messages you when it needs a decision or has finished. A chatbot waits for you to type. An agent comes to you.

Between August and October 2026, five of them launched within weeks of each other. Instinct opened an invite-only beta in August. SpaceXAI put Grok Bot into beta on 11 August. Meta launched Muse on 8 September. OpenAI announced dots at DevDay on 29 September. Anthropic opened Claude Duet as a research preview on 21 October.

If you want the background first, our explainer on what agentic AI means covers the idea. This post is the practical comparison.

Where this comes from. Each product's details are taken from the company's own pages: Instinct's site and terms of service (revised 26 August 2026), TechCrunch's report on Instinct, Meta's Muse announcement, SpaceXAI's Grok Bot announcement, Anthropic's Claude Duet page and safety write-up, and OpenAI's dots documentation. Where we rely on reporting rather than an official page, we say so. These products change weekly, so check the official page before you rely on a price or plan. Last reviewed 10 November 2026.

How do Instinct, Muse, Grok Bot, Claude Duet and ChatGPT dots compare?

Muse's paid-plan prices are as reported at launch rather than taken from Meta's announcement, which says only that Muse is free for most needs and that subscriptions exist for people who want more. Each of the five does its work on its own computer rather than on your device, so your phone or laptop doesn't need to stay on while it works.

What is each personal AI agent best at?

• Instinct: errands by text. You don't install anything. You text or call it, and it uses its own phone number and computer to book trains, order groceries, send flowers or make reservations. TechCrunch reports a concierge feature that phones businesses with no online booking. Access is by invitation, and the company raised $1 billion at a $10 billion valuation in September.
• Muse: everyday life, free. Meta pitches it at long-term goals that it turns into plans and then works on: booking travel, filling in forms, sending emails, even negotiating. It plugs into Instagram (a saved recipe reel can become a grocery list) and lives inside WhatsApp. It runs on Meta's own Muse Spark model, which we covered in our Muse Spark explainer.
• Grok Bot: work across tools. SpaceXAI frames Bots as AI teammates. They sign in to the tools you already use, including ones without an API, learn a routine by watching you do it once, and run it on a schedule. You can run several Bots, including a chief-of-staff Bot that oversees specialists. Our Grok Bot guide goes deeper.
• Claude Duet: a daily brief and admin. Duet sends a morning brief, checks your connected apps every few hours and messages you only when something needs you. Anthropic's examples are mostly small-business admin: unpaid invoices, payroll prep, contracts. It connects to more than 6,000 connectors and plugins. See our Claude Duet explainer.
• ChatGPT dots: ongoing projects. A dot handles ongoing responsibilities between conversations: it tracks progress, runs background tasks in parallel and can hand work to ChatGPT's Work and Codex tools. It can also connect to your own computer if you allow it. It inherits your ChatGPT memory, so it starts out knowing what ChatGPT already knows about you.

How do these agents ask before they act?

This is the part that matters most, and the part the launch videos skip. All five can act in your name. What differs is where the line sits between doing something and asking you first.

• Claude Duet reads and drafts on its own, but shows you exactly what it's about to do and waits for your approval before it sends a message, makes a payment, signs a document or deletes anything. A separate check outside the model judges whether each action matches what you asked for. Some actions, such as trading securities or moving money between accounts, are blocked even if you approve them.
• ChatGPT dots use rules: each kind of action can be set to act without asking, act if pre-approved, ask first, or hand off to you. A built-in review decides which applies, and your own rules can't switch off the built-in safety requirements. Background research is read-only.
• Muse asks before sensitive actions like sending an email or making a purchase. A separate agent Meta calls Sentinel must approve any outbound internet activity, and you get an audit trail of past and planned actions.
• Grok Bot works jobs end to end and comes back when something needs your approval or judgment. SpaceXAI's announcement says less about where that line sits than the others do.
• Instinct's terms appoint it as your agent to enter agreements, commitments or transactions on your behalf, and make you responsible for the consequences of its actions.

None of these gates helps if you approve without reading. Our guide to how AI agents work covers the three habits that keep delegation safe: a clear scope, a gate on risky steps and a check on the output.

What do you hand over when you connect a personal AI agent?

An agent that signs in to your accounts holds your sessions. Before choosing one, check three things on its own pages:

1. Passwords. Anthropic says Claude Duet's model never sees your passwords or sign-in tokens. Meta says Muse can use stored credentials without seeing them. OpenAI says dots use a private sign-in flow that keeps credentials out of the conversation.
2. Training. Instinct's terms allow it to use your materials to train AI models unless you opt out in its settings, and the opt-out only applies going forward. Meta lets you opt out of Muse training and says Muse data isn't shared with its ad systems. For dots, personal plans follow ChatGPT's existing data controls.
3. Track record. These products are weeks old. TechCrunch reported that Instinct's first privacy policy was criticised as overreaching and has since been updated. Reporting summarised on Wikipedia says Amazon blocked Muse from its store in September, and that Meta suspended Muse's shared-chat feature after a security group showed shared chats could be found. Expect more stories like these for every product on this list.

A sensible start with any of them: connect one or two apps, give it reversible jobs such as drafting and sorting, and read every approval in the first weeks.

Which personal AI agent should you pick?

• You want to try one today for free: Muse, if you're in the US.
• You already pay for ChatGPT Pro: set up your included dot first. It already knows your ChatGPT history.
• You already pay for Claude in the US: join the Claude Duet waitlist, especially if your days are email, calendar and business admin.
• Your work lives in tools without good integrations: Grok Bot, which signs in to the tools you already use, including ones without an API.
• You want errands done by text and you have an invite: Instinct, after reading its terms.

Whichever you choose, the skill it asks of you is the same. You stop doing the task and start specifying it, fencing it in and checking the result. People who are good at briefing a colleague get far more from these agents than people who type one vague line and hope.

Iro's AI agents path practises exactly that in 5-minute lessons: when to delegate and when to do it yourself, how much to trust an agent's output, and how to spot a confident wrong answer before it goes out under your name. To see where you stand first, take the AI rank quiz.