Choosing an Always-On AI Agent for Your Organization
Three vendors launched personal AI agents in six weeks, but enterprise controls and data policies lag far behind consumer adoption.
November 18, 2026 · 6 min read

Executive summary
Comparing the 3 leading always-on personal AI agents - Meta Muse, OpenAI Dots, and Claude Duet - which all launch within six weeks.
Key metrics
Key takeaways
- Enterprises must publish personal AI agent policy before OpenAI's 18 November 2026 default access deadline, defining permitted agents, data boundaries, and exception approvers.
- Agent traffic represents a new commercial channel class requiring distinct measurement, access decisions, and conversion modeling separate from human visitor analytics.
- Shadow AI risk is acute because Muse and Duet are consumer products that employees may connect to work email, calendars, and documents without IT involvement.
- Only OpenAI Dots offers a documented enterprise admin framework with role-based access, activity review, and monitoring controls suitable for governed pilots.
- Data residency gaps persist: OpenAI states Enterprise dots have no data or inference residency, and Meta uses Muse queries for model training by default.
Three vendors launched personal AI agents in six weeks, but enterprise controls and data policies lag far behind consumer adoption.
Within roughly six weeks, three major AI vendors launched always-on personal AI agents: Meta Muse on 8 September 2026, OpenAI Dots on 29 September 2026, and Anthropic Claude Duet on 21 October 2026.
Each agent operates on its own cloud computer with a browser, continues working after the user closes the app, and pauses for approval before sensitive actions such as sending messages or making payments.
For enterprises, the arrival of always-on personal AI agents is no longer only a question of how they change website traffic; they affect customer channels, IT and security, workforce productivity, finance and procurement, and legal and compliance obligations simultaneously.
Enterprise Readiness Diverges Sharply Across the Three Agents
The three agents differ most in who they are built for and how far they are ready for managed enterprise use. OpenAI Dots is the only one of the three with a documented enterprise admin framework, and it is off by default in Enterprise workspaces during the beta.
Claude Duet is not yet available on Team or Enterprise plans, and Meta Muse is a consumer product available in the US, Canada, Japan and Singapore. OpenAI states that from 18 November 2026, Enterprise members have access to Dots by default unless the workspace opts out, and billing for enabled dots begins on the same date.
Muse is built to take those everyday tasks off your plate, so you have more time for the people and moments that matter.
Muse offers strong per-user isolation through a Secure VM that isolates agent actions, with a separate Sentinel process approving anything that reaches the internet. Credentials are held in secure storage the agent cannot read, and payments use one-time virtual cards. However, Meta can still access data in the VM, and queries are used for training unless the user opts out.
Claude Duet is the most conservative on actions, requiring approval for payments, message sends, signatures, and deletions, with securities trading and moving money between accounts off-limits even with approval.
OpenAI Dots has the most developed enterprise control plane, with background research set to read-only and Custom Rules determining whether an action runs automatically, runs if pre-approved, asks first, or is handed back to the user.
Agent Traffic Reshapes Customer Channels and Conversion
Agents are becoming a new class of visitor and buyer on digital properties. Muse generates consumer-style journeys that run from search to product page to cart and checkout, using Stripe Link one-time cards for payment. Duet behaves like a task-driven professional buyer, for example chasing unpaid invoices or comparing vendors. Dots revisits sites persistently as part of long-running projects.
Singapore is already live for Muse: since 27 October 2026, Muse connects to Shopee, Lazada, Grab (GrabMart and Grab Dine Out), ShopBack, Klook, Trip.com, Singapore Airlines, and Agoda, with some connector features rolling out in phases.
Access is a commercial decision, not only a technical one. Amazon blocked Muse on 21 September 2026, saying the agent did not identify itself, while Shopify allowed it on its storefronts.
Expect different funnel effects: Muse is likely to lift top-of-funnel and consumer checkout activity, Dots to raise mid-funnel revisits for complex products, and Duet to bring fewer but higher-intent B2B visits such as quotes and invoices.
Approval steps and virtual cards may slow final checkout but can reduce abandonment, and conversion on Dots-driven research depends on whether the user sets a purchase goal.
Security, Workforce, and Financial Exposure Require New Controls
Enterprises should treat agents as non-human actors. A Dot cloud computer does not inherit a member's VPN, browser sign-ins, or device policies, and local computer access is a separate opt-in that is off by default for Enterprise. Credential handling is a shared design strength across all three agents, each designed so the model does not see passwords.
Residual risks remain, including prompt injection from web content and, for Muse, vendor access to data in the user's VM. In October 2026 a security group reported that shared Muse chats could be discovered by a third party, and Meta suspended the sharing feature.
On workforce productivity, Duet sends a morning brief, follows up on quiet threads, and builds dashboards, forecasts, and prototypes. Dots can work in Slack and Teams, where other channel members can see what a dot posts under its own identity.
Accuracy still needs human review: Anthropic acknowledges Duet can misread threads and miss items between checks, and OpenAI states Dots can make mistakes and that consequential work should be reviewed. On finance and procurement, billing for enabled Dots begins on 18 November 2026, either at a flat rate by type of work or by consumption, with tasks delegated to Work or Codex charged separately.
Duet draws on the same usage limits as other Claude products. Duet can check QuickBooks or Stripe for unpaid invoices and draft reminders for approval, and Muse pays through Stripe Link virtual cards. Meta has said it expects to profit from a small fee on transactions.
Legal, Compliance, and Data Residency Gaps Persist
Data use and residency remain unresolved across the three offerings. Meta uses Muse queries for model training by default, which users can disable.
OpenAI states that Enterprise Dots have no data or inference residency. Reporting in September 2026 indicated that some Muse phone calls to businesses were completed by human call-center workers during beta testing, which is relevant to disclosure and privacy reviews.
On auditability, Muse provides a user-level audit trail, and OpenAI provides admin activity and compliance records for Enterprise workspaces. Enterprises should confirm what is retained and for how long before approving use.
Banks are locking out AI Agents and shifting the losses to customers who let them in.Linas Beliūnas, Author of Linas's Newsletter
Recommended Actions for Enterprise Leaders
Enterprise leaders should set policy first, confirming whether their OpenAI workspace will opt out or opt in ahead of 18 November 2026 and publishing guidance on personal agents such as Muse and Duet that connect to work accounts.
A governed pilot should start where controls exist, piloting Dots with a defined group using role-based access while keeping local computer access off, and reviewing Duet for Enterprise once Anthropic makes it available. Agents should be managed as identities with least privilege, registered as non-human actors with limited connected apps and actions.
Approval thresholds should require human sign-off for payments, external messages, signatures, deletions, and any customer or financial data sharing.
Enterprises should set a customer-channel stance on agent traffic, measuring agent-driven traffic and conversion and modeling the impact for their business type.
Risk and finance reviews should be updated to add agents to vendor risk assessments, covering data retention, training use, and residency, and to budget for consumption or flat-rate billing.
In summary, Meta Muse is likely to drive the most consumer traffic and checkout attempts, Claude Duet behaves like a work-focused buyer with cautious approvals, and OpenAI Dots acts as a persistent project agent with the most mature enterprise controls.
The right priorities will depend on sector - whether e-commerce, travel, SaaS, or marketplace - and developing a traffic and conversion impact model alongside governance steps is recommended.
Editorial FAQs
What are always-on personal AI agents and how do they differ from standard chatbots?
Always-on personal AI agents such as Meta Muse, OpenAI Dots, and Anthropic Claude Duet each run on their own isolated cloud computer with a browser, continue working after the user closes the app, and pause for approval before sensitive actions such as sending messages or making payments.
Which of the three personal AI agents is most ready for enterprise deployment?
OpenAI Dots is the only one of the three with a documented enterprise admin framework, offering role-based access, Slack and Teams permissions, cloud and local computer settings, and usage and activity review. It is off by default in Enterprise workspaces during the beta.
What governance actions should enterprise leaders take before November 2026?
Leaders should confirm whether their OpenAI workspace will opt out or opt in ahead of 18 November 2026, publish guidance on personal agents connecting to work accounts, run a governed pilot with role-based access, register agents as non-human identities, and add agents to vendor risk assessments covering data retention, training use, and residency.