Manus 2.0 Launches Cue, Rivalling Meta's Muse in AI Agents

What happened

Manus has released Manus 2.0, a rebuilt version of its AI agent platform that the company describes as a new architecture rather than an incremental update. The release spans web, desktop and mobile, and introduces a new agent framework called Cascade, dedicated execution environments for running tasks, event-triggered automations, an upgraded desktop client called Manus Studio, and Cue, a standalone app for personal AI agents.

The launch is notable for its backdrop: Manus and Meta had reportedly discussed a merger earlier this year before those talks collapsed. With Cue, Manus now finds itself competing directly with Meta's own personal-agent app, Muse, in the emerging market for consumer-facing agentic AI.

According to Manus, Cue lets users run personal agents on phone and desktop, built on the same underlying infrastructure as the core Manus product. Each agent within Cue is designed to operate with its own identity and resources — including a distinct email address, phone number and digital wallet — rather than sharing a single generic assistant profile.

Why it matters

The move signals where agentic AI is heading next: away from single-purpose chat assistants and toward multiple, persistent agents that each hold their own credentials, contact details and even payment means. Giving an agent its own email, phone number and wallet is a meaningful architectural choice — it implies agents are being designed to transact and communicate autonomously on a user's behalf, not just answer prompts inside a chat window.

For organisations tracking digital transformation and AI adoption, this points to a fast-maturing agent ecosystem where platform architecture — execution environments, automation triggers, identity management — is becoming as competitively important as model quality itself. It also underscores how quickly former partners can become rivals once a merger falls through, with both parties racing to define the standard for how personal agents should work.

The Renascence take

The detail most likely to be skipped over — an agent having its own phone number, email and wallet — is actually the story. It reframes what "agentic AI" means in practice: not a smarter chatbot, but a semi-autonomous actor with the means to communicate and pay on someone's behalf.

Handing an agent its own identity and wallet is a trust decision dressed up as a technical feature, and most users will not read it that way until something goes wrong. The behavioural risk isn't capability, it's delegation without visibility — people will grant an agent autonomy long before they've built a mental model of what it can actually commit them to. Any operator building on this kind of infrastructure should treat transparency and reversibility of agent actions as the real product, not the automation itself.