Manus Cue gives every agent an email, phone, and wallet

Something is going to call your business next quarter, and it will not be a person. Manus shipped Manus 2.0 and a new app called Cue on September 28, and the detail that matters for anyone who takes orders or bookings is this: every Cue agent gets its own email address, its own phone number, and its own wallet with a budget you set. It can place calls, take calls and hand you a summary, and pay for things. The agentic commerce conversation just stopped being theoretical about identity.

What actually happened

Cue is a standalone app for personal agents, live on web, desktop, and mobile, with iOS pending App Store review. It is in early access, free with the invite code MEETCUE on a limited, first-come basis. Beyond the email, phone, and wallet, each agent runs on its own machine, works in group chats where you delegate tasks to specific agents, and — per Manus — can scan a restaurant QR code to place an order or make a reservation. Manus's own framing on spend is that the agent "pays within the budget you set," which is a control worth noting because it is the only one described.

The rest of Manus 2.0 is a platform rebuild. A new agent harness called Cascade that Manus reports uses 23.2% fewer tokens, finishes tasks 28.2% faster, and costs 32% less — all vendor-reported, no independent benchmark. A Cloud Computer for projects that need to run around the clock. Automations that fire on events in email, calendar, Slack, or Notion. Manus Studio for documents, spreadsheets, slides, sites, code, and video. Investing.com covered the launch as a platform expansion.

Why agent identity matters for your checkout and your phone line

Strip the product away and you get a factual change: consumer software now holds a phone number, an email address, and a payment instrument, and it will use all three against businesses that never agreed to serve it. Your reservation form, your support inbox, your voice line, your checkout. None of them were designed with a decision about this.

Three questions to settle this quarter, before the volume shows up.

Do you want agent traffic? It is not a rhetorical question. An agent that orders, pays the listed price, and does not negotiate is a good customer for a high-margin product and a terrible one for anything needing a conversation about fit, sizing, or timing. Pick deliberately. A restaurant taking QR-code orders from software has a different answer than a custom fabricator.

Can you tell? Right now, probably not. Log enough to find out: user agent, request headers, call metadata, whether an email thread ever hits a human's reading speed. You cannot make a policy about traffic you cannot identify, and retrofitting that logging after a fraud dispute is the expensive path.

What is your refund policy when an agent got it wrong? An agent with a budget can misread your menu, your options, or your lead time and pay anyway. The chargeback lands on you, the human behind the agent is annoyed, and your terms say nothing about it. Write the clause now.

If the answer to the first question is yes, publish a structured endpoint — clean product and availability data, real prices, honest stock — so agents read a feed instead of scraping a form and guessing. Scraping is where wrong orders come from. And keep the record in your system, not in the agent's transcript. That transcript belongs to somebody else's software on somebody else's roadmap.

Sources: Manus, Investing.com.