Manus drops 2.0 with a new creative studio and autonomous agents app Cue

Anyone who has ever tried making a video or game with AI knows the pain: you get a decent first draft, but if you want to swap out one background track or change a character's hair color, you basically have to reroll the prompt and pray to the RNG gods.

Manus wants to get rid of that slot-machine workflow.

The company rolled out Manus 2.0. Alongside a completely overhauled agent architecture dubbed Cascade, the release turns their desktop client into Manus Studio, brings specialized editors for video and game development, and introduces a standalone personal assistant app called Cue.

Giving AI an actual timeline

Instead of treating agents like opaque black boxes, Manus Studio is moving toward a shared human-AI workspace that only pulls in heavy tools when the task calls for them.

Under the hood, the new Cascade framework reportedly cut token usage by 23.2%, trimmed task completion time by 28.2%, and dropped operational costs by 32% in test setups. It basically stops the system from dragging irrelevant context into every single prompt.

That shows up clearly in the new video editor. Instead of spitting out an uneditable MP4, Manus now hands you an interactive timeline. Clips, audio, motion graphics, and text sit on separate layers. You can manually swap out an AI background track for your own audio, drop in real product photos, or toggle on Alchemy Mode, which pairs video generation with code generation for tighter creative pacing.

Game Dev takes a similar path. Running in Max Ultra mode, you can play a generated build immediately, click on an individual NPC to change asset code, and tweak physics by hand. To solve the hosting nightmare that usually kills browser multiplayer projects, Manus also introduced Cloud Computer: persistent virtual machines you can rent with a few clicks to host game servers or 24/7 automation triggers.

Phone numbers and wallets for your agents

While Studio focuses on creative work, Cue is built entirely for personal life as a standalone mobile and desktop app.

What makes Cue stand out is that every agent gets its own distinct identity. That means its own email address, a working phone number, an isolated computer environment, and a dedicated wallet with spending caps you control.

Cue agents can pick up phone calls, leave meeting summaries, and pay for services autonomously within your budget limits. In offline settings, they can scan QR codes to handle food orders or wait in line. You can also throw multiple agents into a group chat to collaborate on larger projects, like scouting venues and drafting presentation decks for an event. Cue is currently in early preview using the invite code MEETCUE.

Why it matters

We think Manus 2.0 makes a crucial concession: prompt-and-pray automation is hitting a hard ceiling in professional workflows.

Giving creators raw timelines, editable code blocks, and granular asset pickers looks like giving up on pure autonomy, but it is actually what makes the tool usable. Getting 90% of a video done in seconds is useless if fixing the last 10% takes five hours of prompt wrestling. Meeting humans halfway in a proper UI is the right move.

Cue's push to give agents real wallets and telecom identities feels ambitious, but it also carries obvious risks. Handing autonomous software the keys to personal spending and inbound calls will inevitably bump into prompt injection exploits and authorization failures. Still, for creators tired of black-box chatbots, Studio is well worth testing today, and it puts serious pressure on rival agent platforms to build real software interfaces instead of endless chat windows.