The AI assistant that waits for you to type is giving way to one that works while you sleep. On October 21, 2026, Anthropic opened a research preview of Claude Duet, an app in which Claude monitors your accounts, completes assigned tasks in the background and messages you when something needs a decision. It joins OpenAI’s Dots and Meta’s Muse agent in what is turning into a three-way race for the always-on personal agent.
What Duet does
According to Anthropic’s description, Duet sends a short morning brief across connected apps, checks those accounts every few hours, and only messages when something needs attention. Tasks run on its own virtual machine with its own browser, so a laptop or phone need not stay on. It keeps one ongoing thread for typed or voice messages and inherits memory, skills and connected apps from the user’s Claude account. Anthropic says it works with more than 6,000 connectors and plugins, including Gmail, Slack, QuickBooks, Stripe, HubSpot and Shopify, and it is aimed at professionals and small business owners.
The product is built on Claude Opus 5.5, per coverage from eesel, and runs on iOS, Android and claude.ai. It is included in Pro and Max plans and shares their usage limits, so background checks count against the same allowance as ordinary chats. Access is by waitlist and starts with US subscribers.
Where the guardrails sit
The permission model is the most useful part to understand. Duet reads, drafts and prepares work on its own. Before it sends a message, makes a payment, signs a document or deletes anything, it shows the exact action and waits for approval. Some actions are off-limits even with approval, including trading securities and moving money between accounts. Payments go through one-time virtual cards via Stripe’s Link or Shop Pay, and Anthropic says the agent does not directly see passwords or sign-in tokens. A classifier blocks attempts to share the most sensitive data, such as card numbers and government ID, and health or personal finance information is shared only on explicit instruction.
Anthropic is candid about the limits. It says Duet may misread a thread or misjudge urgency, that personal errands such as booking tickets are weaker than work tasks, and that it cannot yet show the user its browser. The eesel write-up adds that no prompt-injection defense is perfect, which Anthropic cites as part of why consequential actions need approval, and notes Microsoft 365 is not yet supported even though a product page shows its logo. Because eesel sells a competing helpdesk product, its comparisons deserve a skeptical read, but the factual details match Anthropic’s own materials.
The competition
Eesel’s comparison table lists OpenAI’s Dots as launched on September 29, 2026 on GPT-6 Astra, available from ChatGPT Pro at $100 a month and working inside ChatGPT, Slack and Teams, and not offered on Pro in the European Economic Area, Switzerland or the UK. Meta’s Muse agent is listed as launched September 8 on Muse Spark 1.3, free to start, reachable through iOS, Android, web and WhatsApp, and US-only. A separate release tracker, ThursdAI, describes Dots as always-on agents with their own cloud computer and browser, restricted to Pro, and notes it was among OpenAI’s DevDay announcements. The launch dates differ slightly between trackers, which is typical for staged rollouts.
The structural differences are visible even through vendor-filtered summaries. Anthropic’s pitch is a single person’s inbox, calendar and side projects; it does not support team plans yet. OpenAI is positioning Dots inside tools that companies already use for work. Meta is leading with a free consumer product and a messaging app that already has a very large user base. Each approach bets on a different answer to the same question: who is the agent working for, and where does the person want to supervise it?
The cost nobody advertises
Always-on means always consuming. Eesel reports that Duet adds no extra fee to Pro or Max, but its checks draw on the same five-hour and weekly limits, and admitted users get one full limit reset usable until November 30, 2026. A user who connects several busy accounts may find the allowance disappearing faster than expected, since larger tasks like financial models or decks use more. That trade is less visible than a monthly price and harder to plan around.
There is a security trade too. An agent with access to email, payment tools and signing services is a more valuable target than a chatbot. The approval prompts only help if the person actually reads them; a stream of routine confirmations can train users to click through. Evaluating these products means asking how often they interrupt, how clearly they show what they are about to do, and how easy it is to revoke access.
What This Means for Philippine Founders
Nobody in the Philippines can simply sign up for Duet yet: the preview is described as US-only and waitlist-based, and Anthropic has not said when other countries are next. Dots availability outside the regions named above is not spelled out in the sources reviewed here, so Philippine availability should be treated as unconfirmed. Founders who want to experiment can watch the waitlists, but should not plan a product or workflow around access they cannot yet verify.
What can be done now is preparation. An always-on agent is only useful if the underlying accounts are tidy: a clean inbox, a shared calendar, documented approval rules for payments and contracts. Teams can also draft a short policy on what an AI agent may read, draft or send on behalf of an officer, and which actions always require a person, mirroring the approve-first pattern Anthropic describes. If a Philippine startup builds its own agent features, the same design choice applies: let the software prepare the work and keep humans on the actions that move money or make commitments.
Share this article