Product1 publisher3 min readPublished
In Manus's new Cue app, each personal agent gets its own wallet and phone number
Manus's rebuilt 2.0 platform adds Cue, an app in which each personal agent has its own email, phone number, wallet and computer. Those agents can spend within set limits. A team that allows Cue at work has to set those limits first.
The Product Desk · Product desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Computerworld describes Cue as Manus's alternative to Meta's Muse app, arriving months after Manus and Meta called off their merger.
- Manus 2.0 runs on web, desktop and mobile, and also ships a new agent harness called Cascade and an upgraded desktop app called Manus Studio.
- Manus now sells a Cloud Computer, a dedicated environment it calls a permanent home for an automation that needs to keep running.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- exposure Customers and suppliers can now get email and calls from an agent's own address and number. An agent's mistake reaches people outside the company without going through any employee's outbox.
- constraint Manus did not describe the tasks behind its Cascade figures. Without them, a buyer cannot use the 32% saving to forecast its own bill.
- cost An automation that has to stay running needs a Cloud Computer, a dedicated machine bought on top of the base service.
The first time a Cue agent pays for something, someone in finance will ask who set its limit. Manus says the agents can "transact within defined limits" [6]. Computerworld's account of the release does not say who sets those limits, or whether they apply to one agent or to a group working toward one goal.
Manus pitches 2.0 as "a new architecture, new products, and new capabilities" [2]. In Cue, an agent communicates through its own email and phone number and pays from its own wallet [6]. Elsewhere in Manus 2.0, work can start with nobody present, when a new email or Slack message arrives or a calendar event comes up [11].
The product also lets several agents in a group chat with a shared goal "hand work to each other" [7], and each of those agents has its own wallet [6]. If caps are set per agent, they do not limit what the group spends on the goal.
The desktop features raise a related question, about seeing versus approving. In an authorized session, Manus "works in its own visible workspace on your computer, using the files, browser, and apps you've approved" [13]. It can keep working after the user goes offline, and the user can see the activity [14]. Visible activity is telemetry. An approval gate holds the action until a person agrees, and Gartner senior director analyst Anushree Verma lists the two as separate requirements.
"The capability is becoming enterprise-relevant, but the autonomy is ahead of the governance," Verma said [15]. She said enterprises should treat such systems as "untrusted or semi-trusted automation with strong isolation, approval gates, detailed telemetry, and narrowly defined authority" [16].
For enterprises, Computerworld points to Cascade, the harness that coordinates agents within a project and brings in specialized capabilities only when they are needed [8]. In the one configuration Manus reported, Cascade used 23.2% fewer tokens and finished tasks in 28.2% less time than the previous system [9]. Cost fell 8.8 points further than token use [1]. Something besides token volume got cheaper. Verma's view is that the strongest systems use a model "selectively rather than indiscriminately" [17].
Cue itself is built for one person's phone and desktop [4]. If employees bring it to work, I'd expect the rules to fall to whoever already answers for their inbox and their company card.
My recommendation is to allow the supervised parts of Manus 2.0 now. Keep Cue wallets empty, or at a small hard cap, until Manus shows how limits work across a group chat. The cost of that choice is speed: an agent that has to ask before it pays gives up the independent task completion Manus is selling [6].
To sort everything else, place each proposed task on two axes. The first is whether it spends money or speaks under the agent's own email or number. The second is whether a person is present while it runs.
- Present, spending nothing: shared work in Manus Studio, which Manus calls "a shared workspace for people and AI" [18]. Low risk. - Unattended, spending nothing: event-triggered automations and offline desktop sessions. These need someone reading the logs, and a short list of approved files. - Present, spending or speaking: a person approves each payment or outside message. - Unattended, spending or speaking: the case Cue is built for. It needs narrow authority and a hard cap per agent and per group, or an empty wallet.
What to watch
- Manus documentation showing who sets Cue spending limits, and whether they apply per agent or per group chat.
- A published price for Cloud Computer, the dedicated environment for automations that must keep running.
- Cascade results on tasks Manus names, or tests by outside parties, measured against the 32% cost figure.