Claude Duet vs Meta Muse vs OpenAI Dots | Sydney AI Guide

Anthropic launched Claude Duet on 21 October 2026 as a proactive, cloud-based AI agent, joining Meta Muse and OpenAI Dots. The three differ in how they surface work, connect applications and supervise consequential actions. For Sydney and NSW businesses, the choice depends less on headline capability than on access, approval controls, data handling and supervision costs. Duet remains a US research preview, not a generally available Australian product.

ELYMENT Insights | Artificial Intelligence & Business Operations | Sydney, NSW | 26 October 2026

Three AI Agents Are Competing for the Same Working Day

Artificial intelligence companies are moving beyond assistants that wait for instructions. Their latest products are designed to notice unfinished work, operate across connected applications and return with results that would otherwise require repeated human intervention.

Anthropic entered this increasingly competitive market on 21 October 2026 with Claude Duet, a standalone AI agent powered by Claude Opus 5.5.

The timing is significant. Meta introduced Muse on 8 September, while OpenAI announced Dots on 29 September.

All three companies are positioning their products as persistent digital assistants with access to cloud-based computers, connected services and continuing context.

But describing them as interchangeable would overlook important differences in their product design, approval mechanisms, access arrangements and commercial maturity.

For Australian organisations, particularly those coordinating customers, suppliers, consultants and operational projects, these differences affect whether an agent is genuinely useful or simply introduces another system requiring supervision.

The central question is no longer whether AI can perform an administrative task. It is which AI product can complete the task accurately, with appropriate permissions, at a justifiable operating cost.

Claude Duet vs Meta Muse vs OpenAI Dots: The Comparison

The three products share the broader concept of autonomous, cloud-based assistance, but their initial emphasis differs.

The comparison below reflects publicly announced capabilities as at 26 October 2026, rather than independent benchmarking of task-completion accuracy.

Developer

• Claude Duet: Anthropic
• Meta Muse: Meta
• OpenAI Dots: OpenAI

Announced

• Claude Duet: 21 October 2026
• Meta Muse: 8 September 2026
• OpenAI Dots: 29 September 2026

Underlying AI

• Claude Duet: Claude Opus 5.5
• Meta Muse: Muse Spark
• OpenAI Dots: GPT-6 Astra

Primary Emphasis

• Claude Duet: Professional prioritisation, daily briefings and delegated work.
• Meta Muse: Personal assistance, consumer tasks and connected business activity.
• OpenAI Dots: Persistent projects, continuing context and multi-step execution.

Execution Environment

• Claude Duet: Dedicated cloud virtual machine.
• Meta Muse: Muse Secure VM.
• OpenAI Dots: Independent cloud computer.

Connected Services

• Claude Duet: Access to more than 6,000 Claude connectors and plugins.
• Meta Muse: Consumer and business integrations, including Meta services.
• OpenAI Dots: Access to an ecosystem of more than 4,000 applications.

Human Controls

• Claude Duet: Approval required for sending, payments, signing and deletion.
• Meta Muse: Approval for sensitive actions, including sending and spending.
• OpenAI Dots: Action review, configurable rules and applicable approval controls.

Initial Availability

• Claude Duet: US Pro and Max research preview via waitlist.
• Meta Muse: Consumer rollout initially in the US, later Canada.
• OpenAI Dots: Gradual rollout across eligible paid and business plans.

The published connector figures are ecosystem claims. They do not establish that every listed application, function or permission is available to every subscriber.

Sources: Anthropic's Claude Duet announcement, Meta's Muse announcement and OpenAI's introduction to Dots.

Anthropic Is Competing for Attention, Not Just Automation

Claude Duet's distinguishing proposition is its attempt to decide which information deserves the user's attention.

According to Anthropic, Duet produces a morning briefing using information from connected applications. It also checks for developments during the day and can undertake delegated work while the user is away.

This includes preparing forecasts, building dashboards, arranging meetings, drafting communications and developing web applications.

Importantly, Anthropic says Duet checks connected accounts every few hours rather than continuously, partly to manage subscription usage limits.

That detail exposes an economic consideration absent from many AI product demonstrations: autonomous monitoring consumes resources even when the user has not initiated a new conversation.

For a Sydney business coordinator responsible for customer correspondence, outstanding invoices, contractor updates and management reporting, the attraction is straightforward.

A useful briefing might identify three issues requiring decisions rather than reproduce 40 incoming messages.

But selective reporting also creates risk. An agent that incorrectly classifies a critical issue as low priority may create an operational delay precisely because its user expects it to filter information intelligently.

Anthropic acknowledges that the research-preview version may misinterpret conversations or incorrectly judge urgency. It also cannot currently expose its browser for users to inspect in the way some competing systems can.

Claude Duet should therefore be evaluated for the quality of its prioritisation, not simply the amount of work it can initiate.

Meta Muse Has a Different Advantage: The Customer Ecosystem

Meta's strategy places considerable emphasis on consumer interactions, messaging and commerce.

Muse was introduced as a personal agent accessible through a dedicated application and WhatsApp, with cloud-based execution, connected services and the ability to act on longer-term goals.

In a subsequent announcement on 29 September 2026, Meta expanded its proposition with Muse for Small Business.

That expansion added access to services including Facebook and Instagram business accounts, Shopify, Stripe, QuickBooks, Canva and several other business applications.

The implications extend beyond personal productivity.

A retailer may value an agent capable of working with its storefront and marketing tools. A service business may see greater benefit from assistance with enquiries, customer communications and campaign preparation.

Meta states that publishing, sending messages and spending require user approval.

Its architecture also includes a separate security system intended to assess actions before they proceed.

However, an integration with a social platform does not establish that the agent understands the commercial obligations behind a conversation.

A customer enquiry about a renovation quotation, for example, might involve access restrictions, scope exclusions or qualifications that are not obvious from the latest message.

Meta's advantage may be the breadth of its customer-facing ecosystem. The business still needs to evaluate the reliability of outputs prepared from that ecosystem.

Elyment has separately examined the implications of Meta Muse becoming a more active digital participant in business communications.

The competitive issue now is whether those capabilities are better suited to a company's existing work than the alternatives.

OpenAI Dots Is Positioning Itself Around Continuing Projects

OpenAI's Dots takes another approach, placing particular emphasis on persistent context, longer-running assignments and continuing relationships between users and their agents.

A Dot can use an independent cloud computer, supported connected applications and contextual information to undertake work across sessions.

OpenAI also describes interaction through ChatGPT, Slack and Microsoft Teams, alongside controls for reviewing activity and determining how the agent may act.

Unlike Duet's current preview, OpenAI provides a way for users to inspect the Dot's cloud computer.

For organisations conducting multi-stage digital projects, this visibility can be an important part of supervision.

Consider a commercial property operator preparing a sequence of project reports for different stakeholders.

Some tasks may involve gathering information, others drafting documents or developing a supporting application.

A persistent agent could support continuity between those activities without requiring the entire project to be reconstructed at every interaction.

Nevertheless, persistent context is not equivalent to an authoritative business record. Users still need to verify source information and review consequential outputs.

OpenAI's controls include approval mechanisms, activity visibility and configurable rules for what an agent may do independently.

Elyment's earlier analysis of how OpenAI Dots maintains work between conversations explores that continuity model in detail.

The decision facing businesses is whether persistent project execution offers more value than proactive prioritisation or customer-facing integration.

The Comparison That Matters: Give All Three the Same Work

Supplier demonstrations are rarely conducted under identical conditions. One product may be shown arranging a meeting, another managing online shopping, and a third producing a sophisticated software application.

These demonstrations reveal capabilities, but they are not reliable evidence of comparative business performance.

Sydney organisations considering AI agents should instead establish a common test using the same records, constraints and expected outcomes.

An Illustrative Sydney Operations Test

Imagine a property management and renovation business coordinating several active jobs. Its administrative workload includes outstanding supplier confirmations, invoice reconciliation and changing customer instructions.

The business prepares a controlled dataset using fictional customer identities, simulated job records and clearly documented instructions.

Each agent is then asked to complete three distinct exercises.

1. Find the unresolved issues. Review a supplied correspondence set, distinguish urgent items from routine updates and provide source references.
2. Prepare the next administrative action. Compare an approved quotation with a revised supplier response and draft a clarification request without sending it.
3. Respect an approval boundary. Prepare a purchasing recommendation involving an expenditure limit, but stop before making any purchase.

The same information and evaluation rules should apply to all three products.

Measure the Work, Not the Demonstration

Accuracy

• What to assess: Were quantities, dates and commercial details reproduced correctly?

Evidence

• What to assess: Could an employee trace conclusions to their original records?

Approval Discipline

• What to assess: Did the agent stop before an unauthorised action?

Interruption Quality

• What to assess: Did it request human attention at the appropriate points?

Recovery

• What to assess: Could an incomplete or incorrect task be reviewed and corrected?

Operating Cost

• What to assess: How much subscription usage, setup and employee supervision did the task require?

This evaluation framework avoids declaring a winner based on the number of connectors, promotional demonstrations or the sophistication of the underlying model.

No independent, like-for-like benchmark establishing a universal winner among these three products is presented here.

The Hidden Cost Is the Work That Still Needs a Person

The subscription fee is only one component of an autonomous agent's real cost.

Business owners also need to consider the effort required to connect systems, validate outputs, correct errors and approve actions.

This can produce unexpected economics.

An agent might finish a draft in minutes but generate several approval requests, each requiring an employee to open the underlying records and check the proposed action.

Another agent may take longer but present a clearer, better-supported result requiring less intervention.

The second outcome could be commercially preferable even if it appears slower.

A meaningful cost assessment should include:

• Subscription and usage allowances.
• Background monitoring and task consumption.
• Application connection and configuration work.
• Time employees spend reviewing outputs.
• Rework caused by incomplete or inaccurate results.
• Records, security administration and ongoing access management.

These factors are particularly relevant to smaller NSW organisations without dedicated teams maintaining automation systems.

Claude Duet currently shares usage limits with the subscriber's broader Claude plan. Meta offers a broadly free consumer entry point with paid options, while OpenAI's first Dot is included in certain eligible paid plans with defined work allowances.

These arrangements are not directly comparable as fixed prices per completed task. Businesses should obtain current regional pricing and usage conditions before procurement.

Australian Privacy Requirements Change the Purchasing Decision

Agentic software presents a distinct information-governance challenge because it may access multiple connected systems and act on the information it encounters.

A Sydney property business could hold customer addresses, tenancy correspondence, payment records, quotations, legal documentation and supplier agreements.

Connecting these sources to an AI agent raises questions about permissions, handling of personal information, international data flows and accountability for external communications.

The Office of the Australian Information Commissioner's guidance on commercial AI products recommends appropriate due diligence, privacy-by-design measures, transparency, ongoing monitoring and human oversight.

Organisations subject to the Privacy Act 1988 must assess their applicable Australian Privacy Principles obligations, including those relating to use, disclosure and protection of personal information.

Smaller businesses should not automatically assume that their size excludes them from every privacy obligation.

Specific legal obligations depend on the organisation, its activities and the information involved.

Before connecting an autonomous agent, decision-makers should establish:

• Which categories of business information the agent genuinely needs.
• Whether existing client disclosures and permissions are appropriate.
• Which applications permit read-only access and which permit changes.
• What records are retained of agent actions and human approvals.
• How access can be suspended or revoked.
• Whether the supplier's regional availability, data handling and contractual terms satisfy business requirements.

A vendor's security architecture is valuable, but it does not replace the customer's responsibility to establish suitable operating procedures.

Which AI Agent Makes Sense for a Sydney Business?

The practical answer depends on the nature of the work being delegated.

Choose to Evaluate Claude Duet When...

The greatest difficulty is managing competing priorities across business systems. Its briefing model and professional-work emphasis make it relevant to administrative coordination, provided the business can access the product and verify performance.

Choose to Evaluate Meta Muse When...

Customer interaction, business messaging, online commerce and Meta's existing ecosystem are central to operations. Its connected-business direction may be attractive to service providers, retailers and customer-facing organisations.

Choose to Evaluate OpenAI Dots When...

The priority is maintaining progress across multi-stage digital projects and providing users with continued context, connected application access and visibility into execution.

These are product-positioning distinctions, not independently proven performance rankings.

Availability may override other factors. Anthropic's Duet research preview started with US Pro and Max subscribers, while Muse's initial rollout focused on North America.

OpenAI's Dots is being introduced gradually across eligible plans and supported markets. Australian businesses should confirm account-specific eligibility before making deployment commitments.

Companies considering broader workflow development can also review Elyment's AI systems and software development services for the distinction between general-purpose AI tools and systems tailored to defined operational requirements.

The Competitive Advantage May Be Fewer Unnecessary Decisions

Claude Duet, Meta Muse and OpenAI Dots illustrate how quickly the AI market is moving from conversation to execution.

Yet the next commercial test is not whether an agent can send more emails, generate more reports or complete more automated steps.

It is whether the agent can remove unnecessary administrative effort while preserving the decisions that genuinely require a person.

In Sydney's property and professional services sectors, the distinction matters because digital work frequently leads to physical, financial or contractual consequences.

A report may influence procurement. An email may create a customer expectation. A calendar update may change contractor attendance. An incorrectly interpreted instruction may cause avoidable rework.

The most valuable AI agent will not necessarily be the one with the broadest promotional feature list.

It will be the one that reliably prepares usable work, communicates uncertainty, respects approval requirements and reduces the organisation's total coordination burden.

As the three products mature, businesses should judge them on verified operational outcomes rather than promised autonomy.

ELYMENT PROPERTY SERVICES

Review Your Next Operational Project With Elyment

Explore practical project planning, workflow coordination, compliance considerations and delivery support for your Sydney operations.

Discuss Your Project With Elyment

Sources and Further Reading

• Anthropic: Claude Duet Announcement
• Meta: Introducing Muse Personal AI Agent
• OpenAI: Introducing Dots
• Office of the Australian Information Commissioner: Guidance on Privacy and the Use of Commercially Available AI Products
• Elyment: Meta Muse AI Agent — From AI Glasses to Digital Employee
• Elyment: Meet OpenAI Dots — The AI Agent That Works After ChatGPT Closes
• Elyment: AI Systems and Software Development Services

Editorial note: Product capabilities, regional availability and subscription conditions reflect published information available on 26 October 2026. Examples of potential business use are illustrative and do not represent independently verified deployments.