Manus 2.0 turns the AI agent into a platform

Manus has released version 2.0 of its AI agent, and the company is positioning the update as a platform shift rather than a routine feature drop. Users can now edit videos, host multiplayer games, and run personal agents remotely from their own phones.

That last detail is the most consequential. Each personal agent gets its own phone number, according to the company. That means a user can delegate tasks to an agent by sending it a message, then step away from the desk entirely. The agent keeps working while the user is offline, and the phone becomes the control surface.

This is the direction the broader agent market has been moving. OpenAI, Google, and Anthropic have all pushed toward agents that complete multi-step tasks autonomously rather than answering single prompts. Manus, which built its reputation on a general-purpose agent that could browse, code, and execute workflows, is now arguing that the next step is infrastructure: give agents persistent capabilities, communication channels, and tools people can share.

The feature list in 2.0 supports that argument. Video editing is a concrete, high-value workload. Multiplayer game hosting is stranger and more interesting — it implies the agent environment can sustain live, stateful services that multiple humans connect to at once. That is closer to a cloud platform than a chatbot.

Consider what these three capabilities have in common. Video editing, game hosting, and phone-reachable agents all require the agent to hold state over time and expose it to other people or systems. A chat interface cannot do that. A platform can. Manus appears to be betting that agents become useful precisely when they stop being conversations and start being services.

The stakes are real. If consumers shift from prompt-based assistants to persistent agents with their own numbers and hosted outputs, the competition moves from model quality to ecosystem breadth — how many real-world tasks an agent platform can actually run end to end. Companies that own that layer could capture significantly more value per user than chatbot makers do today.

Manus faces strong competition. The major labs ship agent products with far larger distribution, and the startup's differentiation has to come from execution speed and the willingness to try unconventional surfaces like game hosting.

Version 2.0 is available now. The open question is adoption: whether users actually assign their agents phone numbers and treat them as always-on services, or whether the platform ambitions outpace what today's models can reliably deliver.