At the end of September 2026 Manus launched Manus 2.0, and the headline feature, Cue, takes a bold position on what an AI agent is. According to Manus's own announcement, each Cue agent gets its own phone number, its own email address, a wallet with a spending cap and an optional cloud computer. It is less a chatbot and more a small digital person with contact details and a payment method.
That is a genuinely different bet from the one behind Claude Code and governed Claude agents. For an Australian business deciding where agentic AI fits, the difference matters more than any feature list.
What is Manus Cue and how is it different from Claude Code?
Manus Cue is an agent identity product: Manus says each agent can take calls and leave summaries, sign up for services under its own email, pay for things within a set cap, join group chats with people and other bots, and act on offline tasks such as scanning a QR code to place an order. Claude Code and Claude agents are built for professional and software work inside systems you control, with scoped permissions, approval gates and audit trails rather than an open-ended identity in the world.
Put simply, Cue gives the agent a presence. Claude gives the agent a job description and a supervisor. Neither is wrong, but they solve different problems.
The details below come from Manus and from launch coverage such as The Next Web's write-up. We have not tested Cue ourselves, so treat the specifics as the vendor's claims.
| Dimension | Manus Cue (per Manus) | Claude Code and Claude agents |
|---|---|---|
| Core idea | Agent with its own identity | Agent doing scoped work in your systems |
| Contact details | Own phone number and email | Acts through connectors you grant |
| Money | Wallet with a spending cap | No wallet; payments go through your approvals |
| Reach | Calls, group chats, offline tasks | Code, documents, business systems via MCP |
| Control model | Caps and settings on the agent | Permissions, approval gates, audit trail |
| Pricing at launch | Phone number about US$9.90 a month, free for a limited time | Your existing Claude plan |
Manus positions Cue against Grok Bot, a standing round-the-clock AI team, and Meta's Muse, a personal-context assistant. Its claimed difference is the real-world identity: an agent with its own contact details and payment method rather than one that acts only inside a chat window.
Why agent identity is a different risk profile
Give an agent its own bank-linked wallet, phone number and email, and let it act semi-autonomously, and you have created something that can make commitments on your behalf. For a consumer booking a restaurant, that may be fine. For a business, the questions arrive quickly:
Who authorised that payment, and can we show it to an auditor?
If the agent agreed to a delivery date on a phone call, is the business bound by it?
Which customer data did the agent share when it signed up for a service under its own email?
How does this sit with Privacy Act obligations if the agent handles personal information in group chats we do not control?
Most business owners we work with do not want an agent with open-ended spending authority. They want defensible, auditable automation: a clear record of who approved a payment or a customer commitment, and an easy way to switch the agent off. We covered a similar split in Claude versus Gemini Spark on always-on agents, and the identity question links to how Claude Tag handles agent identity and access.
What the governed Claude approach looks like in practice
Claude agents built for work tend to follow a simple pattern. The agent connects to your systems through MCP connectors with scoped permissions. Anything that spends money, commits to a customer or changes a record of consequence passes an approval step. Every action is logged. Our guide to approval checkpoints covers where to put the human without slowing everything down.
Take an illustrative example. A Melbourne wholesaler wants an agent to reorder stock. With an identity-style agent, it might place the order itself from its own wallet. With a governed Claude agent, it drafts the purchase order, checks it against agreed supplier pricing and queues it for a buyer to approve. If orders average $8,000 and the business places 40 a month, that is $320,000 of monthly spend. That figure is illustrative, but it shows why a single approval click per order is cheap insurance.
What not to conclude
This is not a verdict that Cue is a bad product. Agent identity is an interesting consumer and prosumer bet, and some of its ideas, such as an agent summarising calls, are useful. Nor does governance mean every action needs a human; low-risk, reversible steps can and should run unattended. The point is narrower: when money, customer commitments or compliance-sensitive records are involved, auditability matters more than giving an agent its own wallet.
If you want automation without giving up control, we scope and build governed Claude agent workflows for Australian businesses. Run the numbers with our ROI calculator or talk to us about where agents fit in your operation.



